diff --git a/AGENTS.md b/AGENTS.md index 1f9714d..ea93660 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -27,15 +27,17 @@ This is where almost all work belongs. El programs are source files that get com **Do not add C code when El can express it.** If functionality can be built from existing El primitives (string ops, `exec`, `fs_read/write`, `http_post`, etc.), write it in El. -### Layer 2: The C runtime (`el-compiler/runtime/el_runtime.c`) +### Layer 2: The C seed (`el-compiler/runtime/el_seed.c`) -This is the hand-written C layer that provides El's native built-in primitives: libcurl HTTP, pthreads server, filesystem I/O, JSON parsing, engram graph store, etc. It is **not generated** — it is maintained by hand. +This is the self-contained C OS-boundary layer. It provides the `__`-prefixed primitives that compiled El programs call: libcurl HTTP, pthreads, filesystem I/O, arena allocation, etc. It is **not generated** — it is maintained by hand. -**Only edit `el_runtime.c` when you genuinely need OS-level access** (raw sockets, GPU calls, new libcurl features). For everything else, write El. +The old `el_runtime.c` has been archived to `el-compiler/runtime/legacy/`. The runtime is now native El (`runtime/*.el`). `el_seed.c` replaces `el_runtime.c` as the sole C compilation dependency. + +**Only edit `el_seed.c` when you genuinely need OS-level access** (raw sockets, GPU calls, new libcurl features). For everything else, write El. When you do add a C builtin: -1. Add the C function to `el_runtime.c` -2. Declare it in `el_runtime.h` +1. Add the C function to `el_seed.c` +2. Declare it in `el_seed.h` 3. Add it to the `builtin_arity` table in `el-compiler/src/codegen.el` (so the compiler knows the arg count) 4. Rebuild the elc binary (see below) @@ -50,14 +52,14 @@ cd /Users/will/Development/neuron-technologies/foundation/el ./dist/platform/elc elc-cli.el > elc-new.c cc -std=c11 -I el-compiler/runtime -lcurl -lpthread \ -o dist/platform/elc-new \ - elc-new.c el-compiler/runtime/el_runtime.c + elc-new.c el-compiler/runtime/el_seed.c # Verify self-hosting: ./dist/platform/elc-new elc-cli.el > elc-verify.c diff elc-new.c elc-verify.c # should be identical mv dist/platform/elc-new dist/platform/elc ``` -After changing `el_runtime.c` only (no El source changes), rebuild downstream programs but do NOT need to rebuild the compiler binary itself — the runtime is linked at the application level, not the compiler level. +After changing `el_seed.c` only (no El source changes), rebuild downstream programs but do NOT need to rebuild the compiler binary itself — the seed is linked at the application level, not the compiler level. --- @@ -66,7 +68,7 @@ After changing `el_runtime.c` only (no El source changes), rebuild downstream pr Each El application has a `build.sh` that: 1. Concatenates all `.el` source files (stripping `import` lines) 2. Runs `elc` to produce a `.c` file -3. Runs `cc` linking against `el_runtime.c` +3. Runs `cc` linking against `el_seed.c` Example (cgi-studio daemon): ```bash @@ -102,8 +104,8 @@ Use `exec()` (blocking) or `exec_bg()` (fire-and-forget) with shell scripts to r | `el-compiler/src/codegen.el` | Code generator — builtin arity table lives here | | `el-compiler/src/lexer.el` | Lexer | | `el-compiler/src/parser.el` | Parser | -| `el-compiler/runtime/el_runtime.c` | Hand-written C runtime (native builtins) | -| `el-compiler/runtime/el_runtime.h` | Runtime header (C function declarations) | +| `el-compiler/runtime/el_seed.c` | Self-contained C OS-boundary layer (replaces el_runtime.c) | +| `el-compiler/runtime/el_seed.h` | Seed header (C function declarations) | | `spec/language.md` | Language specification | | `BOOTSTRAP.md` | How to recover the compiler from scratch | | `elc-cli.el` | Compiler entry point | @@ -126,4 +128,4 @@ exec("EL_HTTP_TIMEOUT_MS=300000 " + SOME_BIN + " " + args + " 2>&1") - New library functions → write in El - New OS/hardware primitives → write in C and register in `codegen.el` arity table - Never edit `dist/platform/elc` directly — always rebuild from source -- Never modify `el_runtime.c` to add functionality that El can express +- Never modify `el_seed.c` to add functionality that El can express diff --git a/el-compiler/runtime/el_runtime.c b/el-compiler/runtime/el_runtime.c deleted file mode 100644 index d091134..0000000 --- a/el-compiler/runtime/el_runtime.c +++ /dev/null @@ -1,10607 +0,0 @@ -/* - * el_runtime.c — El language C runtime implementation - * - * All functions use el_val_t (= int64_t) as the universal value type. - * Strings are transported as their pointer address cast to int64_t. - * On any 64-bit system sizeof(pointer) <= sizeof(int64_t), so this is safe. - * - * Compile with: - * cc -std=c11 -I -lcurl -lpthread -o .c el_runtime.c - * - * Link requirements: -lcurl (HTTP client + LLM), -lpthread (HTTP server). - */ - -/* Feature-test macros must be set before any standard headers. _GNU_SOURCE - * exposes clock_gettime/CLOCK_REALTIME, strcasecmp, and the dlfcn extensions - * (RTLD_DEFAULT) — all of which macOS hands us without asking but glibc on - * Debian gates behind an explicit opt-in. */ -#ifndef _GNU_SOURCE -#define _GNU_SOURCE -#endif - -#include "el_runtime.h" - -#include -#include /* strcasecmp */ -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include /* dlsym for http_set_handler fallback */ -#include -#include -#include -#include -#include -#include - -/* ── Internal allocators ─────────────────────────────────────────────────── */ - -/* - * Per-request string arena - * - * Every El string allocated via el_strbuf / el_strdup during an HTTP request - * is registered in a thread-local arena. When el_request_end() is called at - * the end of the worker thread, every arena entry is freed — recovering all - * the intermediate strings from el_str_concat chains (build_system_prompt, - * engram_compile, etc.) that are otherwise leaked forever. - * - * Long-lived allocations (state_set values, engram internal storage) call - * el_strdup_persist() / el_strbuf_persist() which bypass the arena entirely. - */ - -#define EL_ARENA_INITIAL 512 - -typedef struct { - char** ptrs; - size_t count; - size_t cap; -} ElArena; - -static _Thread_local ElArena _tl_arena = {NULL, 0, 0}; -static _Thread_local int _tl_arena_active = 0; - -/* Binary-safe fs_read length — set by fs_read, consumed by http_send_response. - * Allows serving PNGs and other binary files without strlen truncation. */ -static _Thread_local size_t _tl_fs_read_len = 0; - -static void el_arena_track(char* p) { - if (!_tl_arena_active || !p) return; - if (_tl_arena.count >= _tl_arena.cap) { - size_t nc = _tl_arena.cap == 0 ? EL_ARENA_INITIAL : _tl_arena.cap * 2; - char** grown = realloc(_tl_arena.ptrs, nc * sizeof(char*)); - if (!grown) return; /* can't track — will leak this one ptr, but don't crash */ - _tl_arena.ptrs = grown; - _tl_arena.cap = nc; - } - _tl_arena.ptrs[_tl_arena.count++] = p; -} - -/* Called by http_worker before dispatching the El handler. */ -void el_request_start(void) { - _tl_arena.count = 0; - _tl_arena_active = 1; -} - -/* Called by http_worker after the El handler returns and the response is sent. - * Frees every intermediate string allocated during the request. */ -void el_request_end(void) { - _tl_arena_active = 0; - for (size_t i = 0; i < _tl_arena.count; i++) { - free(_tl_arena.ptrs[i]); - } - _tl_arena.count = 0; -} - -/* Persistent allocation — bypasses the arena (state_set, engram internals). */ -static char* el_strdup_persist(const char* s) { - if (!s) return strdup(""); - return strdup(s); -} -static char* el_strbuf_persist(size_t n) { - char* p = malloc(n + 1); - if (!p) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - p[0] = '\0'; - return p; -} - -static char* el_strdup(const char* s) { - if (!s) { char* p = strdup(""); el_arena_track(p); return p; } - char* p = strdup(s); - el_arena_track(p); - return p; -} - -static char* el_strbuf(size_t n) { - char* p = malloc(n + 1); - if (!p) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - p[0] = '\0'; - el_arena_track(p); - return p; -} - -/* Wrap an allocated C string as el_val_t */ -static el_val_t el_wrap_str(char* s) { - return EL_STR(s); -} - -/* ── I/O ──────────────────────────────────────────────────────────────────── */ - -void println(el_val_t s) { - const char* str = EL_CSTR(s); - if (str) puts(str); - else puts(""); -} - -void print(el_val_t s) { - const char* str = EL_CSTR(s); - if (str) fputs(str, stdout); -} - -el_val_t readline(void) { - char buf[4096]; - if (!fgets(buf, sizeof(buf), stdin)) return el_wrap_str(el_strdup("")); - size_t len = strlen(buf); - if (len > 0 && buf[len - 1] == '\n') buf[len - 1] = '\0'; - return el_wrap_str(el_strdup(buf)); -} - -/* __read_n — read exactly n bytes from stdin. - * Allocates a buffer of size n+1, calls fread(buf, 1, n, stdin) to read - * exactly n raw bytes (including \r, \n, NUL, etc.), null-terminates, and - * returns the buffer as an El String. Returns "" on EOF or I/O error. - * - * Used by the El LSP server to read JSON-RPC message bodies after parsing - * the Content-Length header. readline() cannot be used for the body because - * it stops at the first \n and LSP JSON bodies are not newline-terminated. */ -el_val_t __read_n(el_val_t nv) { - int64_t n = EL_INT(nv); - if (n <= 0) return el_wrap_str(el_strdup("")); - char* buf = malloc((size_t)n + 1); - if (!buf) { fputs("el_runtime: __read_n: out of memory\n", stderr); return el_wrap_str(el_strdup("")); } - size_t got = fread(buf, 1, (size_t)n, stdin); - buf[got] = '\0'; - if (got == 0) { free(buf); return el_wrap_str(el_strdup("")); } - /* Track in arena so the allocation is freed when the request ends. */ - el_arena_track(buf); - return el_wrap_str(buf); -} - -/* __print_raw — write a string to stdout without any modification. - * Unlike println/print (which call puts/fputs and may add newlines or flush - * in platform-specific ways), this uses fwrite with the exact byte count so - * that embedded \r\n pairs in LSP Content-Length headers survive intact. */ -void __print_raw(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return; - size_t len = strlen(s); - fwrite(s, 1, len, stdout); - fflush(stdout); -} - -/* ── String builtins ─────────────────────────────────────────────────────── */ - -el_val_t el_str_concat(el_val_t av, el_val_t bv) { - const char* a = EL_CSTR(av); - const char* b = EL_CSTR(bv); - if (!a) a = ""; - if (!b) b = ""; - size_t la = strlen(a); - size_t lb = strlen(b); - char* out = el_strbuf(la + lb); - memcpy(out, a, la); - memcpy(out + la, b, lb); - out[la + lb] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_eq(el_val_t av, el_val_t bv) { - const char* a = EL_CSTR(av); - const char* b = EL_CSTR(bv); - if (!a || !b) return (el_val_t)(a == b); - return (el_val_t)(strcmp(a, b) == 0); -} - -el_val_t str_starts_with(el_val_t sv, el_val_t prefv) { - const char* s = EL_CSTR(sv); - const char* prefix = EL_CSTR(prefv); - if (!s || !prefix) return 0; - size_t lp = strlen(prefix); - return (el_val_t)(strncmp(s, prefix, lp) == 0); -} - -el_val_t str_ends_with(el_val_t sv, el_val_t sufv) { - const char* s = EL_CSTR(sv); - const char* suffix = EL_CSTR(sufv); - if (!s || !suffix) return 0; - size_t ls = strlen(s); - size_t lsuf = strlen(suffix); - if (lsuf > ls) return 0; - return (el_val_t)(strcmp(s + ls - lsuf, suffix) == 0); -} - -el_val_t str_len(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - return (el_val_t)strlen(s); -} - -el_val_t str_concat(el_val_t a, el_val_t b) { - return el_str_concat(a, b); -} - -el_val_t int_to_str(el_val_t n) { - char buf[32]; - snprintf(buf, sizeof(buf), "%lld", (long long)n); - return el_wrap_str(el_strdup(buf)); -} - -el_val_t str_to_int(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - return (el_val_t)atoll(s); -} - -el_val_t str_slice(el_val_t sv, el_val_t start, el_val_t end) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - int64_t len = (int64_t)strlen(s); - if (start < 0) start = 0; - if (end > len) end = len; - if (start >= end) return el_wrap_str(el_strdup("")); - int64_t sz = end - start; - char* out = el_strbuf((size_t)sz); - memcpy(out, s + start, (size_t)sz); - out[sz] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_contains(el_val_t sv, el_val_t subv) { - const char* s = EL_CSTR(sv); - const char* sub = EL_CSTR(subv); - if (!s || !sub) return 0; - return (el_val_t)(strstr(s, sub) != NULL); -} - -el_val_t str_replace(el_val_t sv, el_val_t fromv, el_val_t tov) { - const char* s = EL_CSTR(sv); - const char* from = EL_CSTR(fromv); - const char* to = EL_CSTR(tov); - if (!s || !from || !to) return el_wrap_str(el_strdup(s ? s : "")); - size_t ls = strlen(s); - size_t lf = strlen(from); - size_t lt = strlen(to); - if (lf == 0) return el_wrap_str(el_strdup(s)); - size_t count = 0; - const char* p = s; - while ((p = strstr(p, from)) != NULL) { count++; p += lf; } - size_t out_sz = ls + count * lt + 1; - char* out = el_strbuf(out_sz); - char* dst = out; - p = s; - const char* found; - while ((found = strstr(p, from)) != NULL) { - size_t chunk = (size_t)(found - p); - memcpy(dst, p, chunk); dst += chunk; - memcpy(dst, to, lt); dst += lt; - p = found + lf; - } - strcpy(dst, p); - return el_wrap_str(out); -} - -el_val_t str_to_upper(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - size_t n = strlen(s); - char* out = el_strbuf(n); - for (size_t i = 0; i < n; i++) out[i] = (char)toupper((unsigned char)s[i]); - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_to_lower(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - size_t n = strlen(s); - char* out = el_strbuf(n); - for (size_t i = 0; i < n; i++) out[i] = (char)tolower((unsigned char)s[i]); - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_trim(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - while (*s && isspace((unsigned char)*s)) s++; - size_t n = strlen(s); - while (n > 0 && isspace((unsigned char)s[n - 1])) n--; - char* out = el_strbuf(n); - memcpy(out, s, n); - out[n] = '\0'; - return el_wrap_str(out); -} - -/* ── Math ────────────────────────────────────────────────────────────────── */ - -el_val_t el_abs(el_val_t n) { return n < 0 ? -n : n; } -el_val_t el_max(el_val_t a, el_val_t b) { return a > b ? a : b; } -el_val_t el_min(el_val_t a, el_val_t b) { return a < b ? a : b; } - -/* ── Refcounted heap objects ────────────────────────────────────────────────── - * - * ElList and ElMap carry a magic-tagged header at offset 0: - * { uint32_t magic; uint32_t refcount; ... payload ... } - * - * The magic tag distinguishes refcounted objects from raw C strings (whose - * first byte is printable ASCII < 0x80) and from small integers (which can't - * be dereferenced). el_retain / el_release sniff the magic and act only on - * matching values; everything else is a safe no-op. - * - * Both ElList and ElMap use INDIRECTION: the header is fixed-size and never - * moves. The payload arrays (elems, keys, values) live in separate heap - * allocations, so realloc-grow on append never invalidates the caller's - * pointer to the header. This is what lets us mutate-in-place safely when - * the refcount is 1 and copy-on-write when it's higher. - * - * Memory model in practice: - * Single-owner accumulator (the cg_stmts pattern) — refcount stays at 1, - * appends amortize to O(1), total memory O(N) for an N-element list. - * Multi-owner branching (the cg_if_stmt pattern) — refcount > 1, each - * append on a shared list copies, so the original is preserved for the - * else-branch. Persistent semantics where they're needed; mutation where - * they're not. */ - -#define EL_MAGIC_LIST 0xE15710A1u /* >= 0x80 in MSB so 'looks_like_string' rejects */ -#define EL_MAGIC_MAP 0xE19A704Bu - -typedef struct { - uint32_t magic; - uint32_t refcount; -} ElHeader; - -/* ── List ────────────────────────────────────────────────────────────────── */ - -typedef struct { - ElHeader hdr; - int64_t length; - int64_t capacity; - el_val_t* elems; -} ElList; - -static ElList* list_alloc(int64_t cap) { - if (cap < 4) cap = 4; - ElList* lst = malloc(sizeof(ElList)); - if (!lst) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - lst->hdr.magic = EL_MAGIC_LIST; - lst->hdr.refcount = 1; - lst->length = 0; - lst->capacity = cap; - lst->elems = malloc((size_t)cap * sizeof(el_val_t)); - if (!lst->elems) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - return lst; -} - -el_val_t el_list_empty(void) { - return EL_STR(list_alloc(4)); -} - -el_val_t el_list_new(el_val_t count, ...) { - ElList* lst = list_alloc(count > 0 ? count : 4); - va_list ap; - va_start(ap, count); - for (int64_t i = 0; i < count; i++) { - lst->elems[i] = va_arg(ap, el_val_t); - } - va_end(ap); - lst->length = count; - return EL_STR(lst); -} - -el_val_t el_list_len(el_val_t listv) { - ElList* lst = (ElList*)(uintptr_t)listv; - if (!lst) return 0; - return lst->length; -} - -el_val_t el_list_get(el_val_t listv, el_val_t index) { - ElList* lst = (ElList*)(uintptr_t)listv; - if (!lst) return 0; - if (index < 0 || index >= lst->length) return 0; - return lst->elems[index]; -} - -el_val_t el_list_append(el_val_t listv, el_val_t elem) { - ElList* old = (ElList*)(uintptr_t)listv; - if (!old) { - ElList* fresh = list_alloc(4); - fresh->elems[0] = elem; - fresh->length = 1; - return EL_STR(fresh); - } - - /* Uniquely owned: grow the elems buffer in place. The header pointer the - * caller holds doesn't move (we only realloc the inner array). This is - * the common case in compiler accumulators, and it's amortized O(1). */ - if (old->hdr.refcount <= 1) { - if (old->length >= old->capacity) { - int64_t new_cap = old->capacity > 0 ? old->capacity * 2 : 4; - el_val_t* grown = realloc(old->elems, (size_t)new_cap * sizeof(el_val_t)); - if (!grown) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - old->elems = grown; - old->capacity = new_cap; - } - old->elems[old->length++] = elem; - return listv; - } - - /* Shared: copy-on-write. The original is preserved for its other owners. */ - int64_t new_cap = old->length + 1; - if (new_cap < 4) new_cap = 4; - ElList* fresh = malloc(sizeof(ElList)); - if (!fresh) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - fresh->hdr.magic = EL_MAGIC_LIST; - fresh->hdr.refcount = 1; - fresh->length = old->length + 1; - fresh->capacity = new_cap; - fresh->elems = malloc((size_t)new_cap * sizeof(el_val_t)); - if (!fresh->elems) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - if (old->length > 0) { - memcpy(fresh->elems, old->elems, (size_t)old->length * sizeof(el_val_t)); - } - fresh->elems[old->length] = elem; - return EL_STR(fresh); -} - -el_val_t el_list_clone(el_val_t listv) { - /* Shallow copy: the new ElList owns its own header and elems buffer, but - * the elements themselves are shared (which is what callers want for the - * cg_if_stmt 'declared' pattern — cloning the spine, not its contents). - * Used by codegen at scope branch points where two child scopes need to - * see the same starting set of declared names without each other's - * mutations. */ - ElList* old = (ElList*)(uintptr_t)listv; - if (!old) return el_list_empty(); - int64_t cap = old->capacity > 0 ? old->capacity : 4; - if (cap < old->length) cap = old->length; - if (cap < 4) cap = 4; - ElList* fresh = malloc(sizeof(ElList)); - if (!fresh) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - fresh->hdr.magic = EL_MAGIC_LIST; - fresh->hdr.refcount = 1; - fresh->length = old->length; - fresh->capacity = cap; - fresh->elems = malloc((size_t)cap * sizeof(el_val_t)); - if (!fresh->elems) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - if (old->length > 0) { - memcpy(fresh->elems, old->elems, (size_t)old->length * sizeof(el_val_t)); - } - return EL_STR(fresh); -} - -/* ── Map ─────────────────────────────────────────────────────────────────── */ - -typedef struct { - ElHeader hdr; - int64_t count; - int64_t capacity; - el_val_t* keys; - el_val_t* values; -} ElMap; - -static ElMap* map_alloc(int64_t cap) { - if (cap < 4) cap = 4; - ElMap* m = malloc(sizeof(ElMap)); - if (!m) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - m->hdr.magic = EL_MAGIC_MAP; - m->hdr.refcount = 1; - m->count = 0; - m->capacity = cap; - m->keys = malloc((size_t)cap * sizeof(el_val_t)); - m->values = malloc((size_t)cap * sizeof(el_val_t)); - if (!m->keys || !m->values) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - return m; -} - -el_val_t el_map_new(el_val_t pair_count, ...) { - ElMap* m = map_alloc(pair_count > 0 ? pair_count : 4); - va_list ap; - va_start(ap, pair_count); - for (int64_t i = 0; i < pair_count; i++) { - m->keys[i] = va_arg(ap, el_val_t); - m->values[i] = va_arg(ap, el_val_t); - } - va_end(ap); - m->count = pair_count; - return EL_STR(m); -} - -static ElMap* as_map(el_val_t v) { return (ElMap*)(uintptr_t)v; } - -el_val_t el_map_get(el_val_t mapv, el_val_t keyv) { - ElMap* m = as_map(mapv); - const char* key = EL_CSTR(keyv); - if (!m || !key) return 0; - for (int64_t i = 0; i < m->count; i++) { - const char* k = EL_CSTR(m->keys[i]); - if (k && strcmp(k, key) == 0) return m->values[i]; - } - return 0; -} - -el_val_t el_get_field(el_val_t mapv, el_val_t keyv) { - return el_map_get(mapv, keyv); -} - -/* Internal: in-place set on a uniquely-owned map. */ -static el_val_t map_set_in_place(ElMap* m, el_val_t keyv, el_val_t value) { - const char* key = EL_CSTR(keyv); - if (key) { - for (int64_t i = 0; i < m->count; i++) { - const char* k = EL_CSTR(m->keys[i]); - if (k && strcmp(k, key) == 0) { m->values[i] = value; return EL_STR(m); } - } - } - if (m->count >= m->capacity) { - int64_t new_cap = m->capacity > 0 ? m->capacity * 2 : 4; - el_val_t* gk = realloc(m->keys, (size_t)new_cap * sizeof(el_val_t)); - el_val_t* gv = realloc(m->values, (size_t)new_cap * sizeof(el_val_t)); - if (!gk || !gv) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - m->keys = gk; - m->values = gv; - m->capacity = new_cap; - } - m->keys[m->count] = keyv; - m->values[m->count] = value; - m->count++; - return EL_STR(m); -} - -el_val_t el_map_set(el_val_t mapv, el_val_t keyv, el_val_t value) { - ElMap* m = as_map(mapv); - if (!m) return 0; - if (m->hdr.refcount <= 1) { - return map_set_in_place(m, keyv, value); - } - /* Shared: copy then set. The original is preserved for its other owners. */ - int64_t new_cap = m->count + 1; - if (new_cap < 4) new_cap = 4; - ElMap* fresh = malloc(sizeof(ElMap)); - if (!fresh) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - fresh->hdr.magic = EL_MAGIC_MAP; - fresh->hdr.refcount = 1; - fresh->count = m->count; - fresh->capacity = new_cap; - fresh->keys = malloc((size_t)new_cap * sizeof(el_val_t)); - fresh->values = malloc((size_t)new_cap * sizeof(el_val_t)); - if (!fresh->keys || !fresh->values) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - if (m->count > 0) { - memcpy(fresh->keys, m->keys, (size_t)m->count * sizeof(el_val_t)); - memcpy(fresh->values, m->values, (size_t)m->count * sizeof(el_val_t)); - } - return map_set_in_place(fresh, keyv, value); -} - -/* ── Refcount ops ─────────────────────────────────────────────────────────── */ -/* - * Both retain and release sniff the magic header to decide whether a value - * is a refcounted heap object. For small integers, raw C strings, and any - * value whose magic word doesn't match, both functions are no-ops. This lets - * codegen emit them on every let-binding without having to track types. - * - * Safety: we filter out obvious non-pointers (small magnitudes, misaligned - * addresses) before dereferencing. For any value that passes the filter and - * lives in a mapped page, reading the first 4 bytes is safe — strings start - * with printable ASCII (< 0x80), so their magic word will never collide with - * EL_MAGIC_LIST (0xE1...) or EL_MAGIC_MAP (0xE1...). Random integers that - * happen to look like aligned heap pointers are exceedingly unlikely to land - * on a page whose first 4 bytes match either magic. */ - -static int looks_like_heap_obj(el_val_t v) { - if (v == 0) return 0; - int64_t s = (int64_t)v; - if (s > -0x10000 && s < 0x10000) return 0; /* small ints */ - uintptr_t p = (uintptr_t)v; - if (p < 0x10000) return 0; /* low addresses */ - if (p & 0x7) return 0; /* malloc returns 8-aligned */ - return 1; -} - -void el_retain(el_val_t v) { - if (!looks_like_heap_obj(v)) return; - ElHeader* h = (ElHeader*)(uintptr_t)v; - if (h->magic == EL_MAGIC_LIST || h->magic == EL_MAGIC_MAP) { - h->refcount++; - } -} - -void el_release(el_val_t v) { - if (!looks_like_heap_obj(v)) return; - ElHeader* h = (ElHeader*)(uintptr_t)v; - if (h->magic == EL_MAGIC_LIST) { - if (h->refcount > 0 && --h->refcount == 0) { - ElList* l = (ElList*)h; - free(l->elems); - l->hdr.magic = 0; /* poison so use-after-free is detected */ - free(l); - } - } else if (h->magic == EL_MAGIC_MAP) { - if (h->refcount > 0 && --h->refcount == 0) { - ElMap* m = (ElMap*)h; - free(m->keys); - free(m->values); - m->hdr.magic = 0; - free(m); - } - } -} - -/* ── Batch 2/3 forward decls (defined later in JSON section) ────────────── */ - -typedef struct JsonBuf JsonBuf; -typedef struct JsonParser JsonParser; -static void jb_init(JsonBuf* b); -static void jb_putc(JsonBuf* b, char c); -static void jb_puts(JsonBuf* b, const char* s); -static void jb_emit_escaped(JsonBuf* b, const char* s); -static int looks_like_string(el_val_t v); -static const char* json_find_key(const char* s, const char* key); -static const char* json_skip_value(const char* p); -static char* jp_parse_string_raw(JsonParser* jp); - -/* Struct definitions are visible here because batch 2/3 helpers above use - * them by value; the bodies (jb_init, etc.) appear in the JSON section. */ -struct JsonBuf { - char* buf; - size_t len; - size_t cap; -}; - -struct JsonParser { - const char* p; - const char* end; - int err; -}; - -/* ── Batch 2: Real HTTP (libcurl client + POSIX-socket server) ───────────── */ -/* - * Client: blocking libcurl easy-handle calls. Errors are returned as a JSON - * fragment {"error":"..."} so callers can detect via str_starts_with("{") / - * json_get_string("error", ...). - * - * Server: bind/listen/accept loop on a TCP socket. Each accepted connection - * is handled in its own pthread (detached). A semaphore-style counter caps - * concurrent in-flight connections at HTTP_MAX_CONNS (64). When the cap is - * reached, accept() blocks until a worker exits. This prevents runaway - * thread creation under high load. - * - * Handler dispatch: El does not expose first-class function references at - * the runtime layer, so the second argument to http_serve(port, handler) is - * treated as a string name (or any el_val_t — the runtime ignores its - * value and uses the registry). Callers register a C-level handler via - * - * extern void el_runtime_register_handler(const char* name, - * el_val_t (*fn)(el_val_t, - * el_val_t, - * el_val_t)); - * - * and select the active handler by calling http_set_handler("name") from - * El, or by setting it directly through the C registry. If no handler is - * registered, the server replies with a 200 carrying a default message so - * the loop is observable. - */ - -/* ── HTTP client write-callback buffer ───────────────────────────────────── */ - -typedef struct { - char* data; - size_t len; - size_t cap; -} HttpBuf; - -static void httpbuf_init(HttpBuf* b) { - b->cap = 1024; - b->len = 0; - b->data = malloc(b->cap); - if (!b->data) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - b->data[0] = '\0'; -} - -static void httpbuf_append(HttpBuf* b, const void* src, size_t n) { - if (b->len + n + 1 > b->cap) { - while (b->len + n + 1 > b->cap) b->cap *= 2; - b->data = realloc(b->data, b->cap); - if (!b->data) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - } - memcpy(b->data + b->len, src, n); - b->len += n; - b->data[b->len] = '\0'; -} - -static size_t http_write_cb(char* ptr, size_t size, size_t nmemb, void* ud) { - size_t n = size * nmemb; - httpbuf_append((HttpBuf*)ud, ptr, n); - return n; -} - -/* JSON-escape an arbitrary C string into an allocated buffer. */ -static char* json_escape_alloc(const char* s) { - if (!s) return el_strdup(""); - JsonBuf b; jb_init(&b); - for (const char* p = s; *p; p++) { - unsigned char c = (unsigned char)*p; - switch (c) { - case '"': jb_puts(&b, "\\\""); break; - case '\\': jb_puts(&b, "\\\\"); break; - case '\n': jb_puts(&b, "\\n"); break; - case '\r': jb_puts(&b, "\\r"); break; - case '\t': jb_puts(&b, "\\t"); break; - default: - if (c < 0x20) { - char tmp[8]; snprintf(tmp, sizeof(tmp), "\\u%04x", c); - jb_puts(&b, tmp); - } else jb_putc(&b, (char)c); - } - } - return b.buf; -} - -static el_val_t http_error_json(const char* msg) { - char* esc = json_escape_alloc(msg ? msg : "unknown error"); - char* buf = el_strbuf(strlen(esc) + 16); - sprintf(buf, "{\"error\":\"%s\"}", esc); - free(esc); - return el_wrap_str(buf); -} - -/* HTTP timeout (ms) — read once from EL_HTTP_TIMEOUT_MS, default 60000. - * Applied via CURLOPT_TIMEOUT_MS on every libcurl request. */ -static long _el_http_timeout_ms = -1; -static long el_http_timeout_ms(void) { - long v = __atomic_load_n(&_el_http_timeout_ms, __ATOMIC_ACQUIRE); - if (v >= 0) return v; - const char* s = getenv("EL_HTTP_TIMEOUT_MS"); - long parsed = 60000L; - if (s && *s) { - char* end = NULL; - long n = strtol(s, &end, 10); - if (end != s && n > 0) parsed = n; - } - __atomic_store_n(&_el_http_timeout_ms, parsed, __ATOMIC_RELEASE); - return parsed; -} - -/* Internal: do a libcurl request; takes optional body/headers, optional method override. */ -static el_val_t http_do(const char* method, const char* url, const char* body, - struct curl_slist* extra_headers) { - if (!url || !*url) return http_error_json("empty url"); - CURL* c = curl_easy_init(); - if (!c) return http_error_json("curl_easy_init failed"); - HttpBuf rb; httpbuf_init(&rb); - char errbuf[CURL_ERROR_SIZE]; errbuf[0] = '\0'; - curl_easy_setopt(c, CURLOPT_URL, url); - curl_easy_setopt(c, CURLOPT_WRITEFUNCTION, http_write_cb); - curl_easy_setopt(c, CURLOPT_WRITEDATA, &rb); - curl_easy_setopt(c, CURLOPT_FOLLOWLOCATION, 1L); - curl_easy_setopt(c, CURLOPT_TIMEOUT_MS, el_http_timeout_ms()); - curl_easy_setopt(c, CURLOPT_NOSIGNAL, 1L); - curl_easy_setopt(c, CURLOPT_ERRORBUFFER, errbuf); - curl_easy_setopt(c, CURLOPT_USERAGENT, "el-runtime/1.0"); - if (extra_headers) curl_easy_setopt(c, CURLOPT_HTTPHEADER, extra_headers); - if (method && strcmp(method, "POST") == 0) { - curl_easy_setopt(c, CURLOPT_POST, 1L); - curl_easy_setopt(c, CURLOPT_POSTFIELDS, body ? body : ""); - curl_easy_setopt(c, CURLOPT_POSTFIELDSIZE, (long)(body ? strlen(body) : 0)); - } else if (method && strcmp(method, "DELETE") == 0) { - curl_easy_setopt(c, CURLOPT_CUSTOMREQUEST, "DELETE"); - } - CURLcode rc = curl_easy_perform(c); - curl_easy_cleanup(c); - if (rc != CURLE_OK) { - free(rb.data); - const char* m = errbuf[0] ? errbuf : curl_easy_strerror(rc); - return http_error_json(m); - } - return el_wrap_str(rb.data); -} - -el_val_t http_get(el_val_t url) { - return http_do("GET", EL_CSTR(url), NULL, NULL); -} - -el_val_t http_post(el_val_t url, el_val_t body) { - return http_do("POST", EL_CSTR(url), EL_CSTR(body), NULL); -} - -el_val_t http_post_json(el_val_t url, el_val_t json_body) { - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - el_val_t r = http_do("POST", EL_CSTR(url), EL_CSTR(json_body), h); - curl_slist_free_all(h); - return r; -} - -/* Build a curl_slist from an ElMap of name -> value strings. */ -static struct curl_slist* headers_from_map(el_val_t headers_map) { - struct curl_slist* h = NULL; - ElMap* m = as_map(headers_map); - if (!m) return NULL; - for (int64_t i = 0; i < m->count; i++) { - const char* k = EL_CSTR(m->keys[i]); - const char* v = EL_CSTR(m->values[i]); - if (!k || !v) continue; - size_t n = strlen(k) + strlen(v) + 4; - char* line = malloc(n); - if (!line) continue; - snprintf(line, n, "%s: %s", k, v); - h = curl_slist_append(h, line); - free(line); - } - return h; -} - -el_val_t http_get_with_headers(el_val_t url, el_val_t headers_map) { - struct curl_slist* h = headers_from_map(headers_map); - el_val_t r = http_do("GET", EL_CSTR(url), NULL, h); - if (h) curl_slist_free_all(h); - return r; -} - -el_val_t http_post_with_headers(el_val_t url, el_val_t body, el_val_t headers_map) { - struct curl_slist* h = headers_from_map(headers_map); - el_val_t r = http_do("POST", EL_CSTR(url), EL_CSTR(body), h); - if (h) curl_slist_free_all(h); - return r; -} - -el_val_t http_post_form_auth(el_val_t url, el_val_t form_body, el_val_t auth_header) { - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/x-www-form-urlencoded"); - const char* a = EL_CSTR(auth_header); - if (a && *a) { - size_t n = strlen(a) + 32; - char* line = malloc(n); - snprintf(line, n, "Authorization: %s", a); - h = curl_slist_append(h, line); - free(line); - } - el_val_t r = http_do("POST", EL_CSTR(url), EL_CSTR(form_body), h); - curl_slist_free_all(h); - return r; -} - -/* HTTP DELETE — mirrors http_post but with CURLOPT_CUSTOMREQUEST=DELETE. - * Returns response body on success; on transport failure returns an error - * JSON fragment (same convention as http_get/http_post). Callers that - * expect "" on failure should check for a leading '{' and an "error" key. */ -el_val_t http_delete(el_val_t url) { - return http_do("DELETE", EL_CSTR(url), NULL, NULL); -} - -/* ── HTTP → file streaming ──────────────────────────────────────────────── - * - * Why this exists: el_val_t strings are NUL-terminated by convention, so - * accumulating an HTTP response into an httpbuf and then wrapping its - * `.data` pointer with el_wrap_str() loses the byte length. Any consumer - * that does strlen() on the wrapped pointer truncates the body at the - * first embedded NUL. Audio (MP3, WAV, OGG), images (PNG, JPEG), and any - * other binary payload hits this. The vessels that download such bodies - * (e.g. ElevenLabs TTS → MP3) get silently corrupted files. - * - * The fix: wire libcurl's CURLOPT_WRITEFUNCTION directly to fwrite() - * against a fopen()-ed FILE*. The bytes never pass through an el_val_t - * string, so embedded NULs are preserved verbatim. Caller's contract is - * just "a file at this path with the response body in it". */ - -static size_t http_file_write_cb(char* ptr, size_t size, size_t nmemb, void* ud) { - FILE* f = (FILE*)ud; - return fwrite(ptr, size, nmemb, f); -} - -/* Internal: stream body to file. method is "GET" or "POST". body may be NULL - * (GET) or NUL-terminated (POST). headers may be NULL. Returns 1/0. */ -static el_val_t http_do_to_file(const char* method, const char* url, - const char* body, struct curl_slist* extra_headers, - const char* output_path) { - if (!url || !*url) return 0; - if (!output_path || !*output_path) return 0; - FILE* f = fopen(output_path, "wb"); - if (!f) return 0; - - CURL* c = curl_easy_init(); - if (!c) { fclose(f); remove(output_path); return 0; } - - char errbuf[CURL_ERROR_SIZE]; errbuf[0] = '\0'; - curl_easy_setopt(c, CURLOPT_URL, url); - curl_easy_setopt(c, CURLOPT_WRITEFUNCTION, http_file_write_cb); - curl_easy_setopt(c, CURLOPT_WRITEDATA, f); - curl_easy_setopt(c, CURLOPT_FOLLOWLOCATION, 1L); - curl_easy_setopt(c, CURLOPT_TIMEOUT_MS, el_http_timeout_ms()); - curl_easy_setopt(c, CURLOPT_NOSIGNAL, 1L); - curl_easy_setopt(c, CURLOPT_ERRORBUFFER, errbuf); - curl_easy_setopt(c, CURLOPT_USERAGENT, "el-runtime/1.0"); - curl_easy_setopt(c, CURLOPT_FAILONERROR, 1L); /* 4xx/5xx → CURLE_HTTP_RETURNED_ERROR */ - if (extra_headers) curl_easy_setopt(c, CURLOPT_HTTPHEADER, extra_headers); - - if (method && strcmp(method, "POST") == 0) { - curl_easy_setopt(c, CURLOPT_POST, 1L); - curl_easy_setopt(c, CURLOPT_POSTFIELDS, body ? body : ""); - /* For the request body we still rely on strlen — POST bodies are - * caller-controlled and JSON/text in every known El use case. - * If a future caller needs a binary POST body, add a *_bytes - * variant that takes an explicit length, mirroring fs_write_bytes. */ - curl_easy_setopt(c, CURLOPT_POSTFIELDSIZE, (long)(body ? strlen(body) : 0)); - } - - CURLcode rc = curl_easy_perform(c); - curl_easy_cleanup(c); - - /* Flush + close before signalling success, so the file is fully on disk - * by the time the caller reads back. */ - int flush_ok = (fflush(f) == 0); - int close_ok = (fclose(f) == 0); - - if (rc != CURLE_OK || !flush_ok || !close_ok) { - remove(output_path); - return 0; - } - return 1; -} - -el_val_t http_get_to_file(el_val_t url, el_val_t headers_map, el_val_t output_path) { - struct curl_slist* h = headers_from_map(headers_map); - el_val_t r = http_do_to_file("GET", EL_CSTR(url), NULL, h, EL_CSTR(output_path)); - if (h) curl_slist_free_all(h); - return r; -} - -el_val_t http_post_to_file(el_val_t url, el_val_t body, el_val_t headers_map, el_val_t output_path) { - struct curl_slist* h = headers_from_map(headers_map); - el_val_t r = http_do_to_file("POST", EL_CSTR(url), EL_CSTR(body), h, EL_CSTR(output_path)); - if (h) curl_slist_free_all(h); - return r; -} - -/* ── HTTP server (POSIX sockets + pthreads) ──────────────────────────────── */ - -#define HTTP_MAX_CONNS 64 - -typedef el_val_t (*http_handler_fn)(el_val_t method, el_val_t path, el_val_t body); - -typedef struct { - char* name; - http_handler_fn fn; -} HttpHandlerEntry; - -static HttpHandlerEntry _http_handlers[32]; -static size_t _http_handler_count = 0; -static char* _http_active_handler = NULL; -static pthread_mutex_t _http_handler_mu = PTHREAD_MUTEX_INITIALIZER; - -static pthread_mutex_t _http_conn_mu = PTHREAD_MUTEX_INITIALIZER; -static pthread_cond_t _http_conn_cv = PTHREAD_COND_INITIALIZER; -static int _http_conn_active = 0; - -/* Public C-level API: register a handler by name. Programs that want El - * `http_serve` to dispatch into their handler call this from main() before - * http_serve. Not declared in the header to keep the public API minimal — - * extern lookup works since C symbols are global. */ -void el_runtime_register_handler(const char* name, http_handler_fn fn); -void el_runtime_register_handler(const char* name, http_handler_fn fn) { - if (!name || !fn) return; - pthread_mutex_lock(&_http_handler_mu); - for (size_t i = 0; i < _http_handler_count; i++) { - if (strcmp(_http_handlers[i].name, name) == 0) { - _http_handlers[i].fn = fn; - pthread_mutex_unlock(&_http_handler_mu); - return; - } - } - if (_http_handler_count < sizeof(_http_handlers) / sizeof(_http_handlers[0])) { - _http_handlers[_http_handler_count].name = el_strdup(name); - _http_handlers[_http_handler_count].fn = fn; - _http_handler_count++; - } - pthread_mutex_unlock(&_http_handler_mu); -} - -void http_set_handler(el_val_t name) { - const char* n = EL_CSTR(name); - pthread_mutex_lock(&_http_handler_mu); - free(_http_active_handler); - _http_active_handler = el_strdup(n ? n : ""); - /* If the name is not yet in the registry, try dlsym lookup against - * the running binary's symbol table. Every El `fn name(...)` compiles - * to a global C symbol with that exact name, so El programs can self- - * register their own handlers just by calling http_set_handler("name"). */ - if (n && *n) { - int found = 0; - for (size_t i = 0; i < _http_handler_count; i++) { - if (strcmp(_http_handlers[i].name, n) == 0) { found = 1; break; } - } - if (!found) { - void* sym = dlsym(RTLD_DEFAULT, n); - if (sym && _http_handler_count < sizeof(_http_handlers) / sizeof(_http_handlers[0])) { - _http_handlers[_http_handler_count].name = el_strdup(n); - _http_handlers[_http_handler_count].fn = (http_handler_fn)sym; - _http_handler_count++; - } - } - } - pthread_mutex_unlock(&_http_handler_mu); -} - -static http_handler_fn http_lookup_active(void) { - http_handler_fn out = NULL; - pthread_mutex_lock(&_http_handler_mu); - if (_http_active_handler) { - for (size_t i = 0; i < _http_handler_count; i++) { - if (strcmp(_http_handlers[i].name, _http_active_handler) == 0) { - out = _http_handlers[i].fn; break; - } - } - } - pthread_mutex_unlock(&_http_handler_mu); - return out; -} - -/* Auto-detect Content-Type from response body. */ -static const char* http_detect_content_type(const char* body) { - if (!body) return "text/plain; charset=utf-8"; - const char* p = body; - /* Binary magic bytes — check before stripping whitespace */ - if ((unsigned char)p[0] == 0x89 && p[1]=='P' && p[2]=='N' && p[3]=='G') - return "image/png"; - if ((unsigned char)p[0] == 0xFF && (unsigned char)p[1] == 0xD8) - return "image/jpeg"; - if (strncmp(p, "GIF8", 4) == 0) return "image/gif"; - if (strncmp(p, "RIFF", 4) == 0) return "image/webp"; - if (strncmp(p, "wOFF", 4) == 0) return "font/woff"; - if (strncmp(p, "wOF2", 4) == 0) return "font/woff2"; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (strncasecmp(p, "= cap) { - if (cap >= 1024 * 1024) { free(buf); return -1; } - cap *= 2; - buf = realloc(buf, cap); - if (!buf) return -1; - } - ssize_t n = recv(fd, buf + len, cap - len - 1, 0); - if (n <= 0) { free(buf); return -1; } - len += (size_t)n; - buf[len] = '\0'; - if (strstr(buf, "\r\n\r\n")) break; - } - /* Parse request line */ - char* sp1 = strchr(buf, ' '); - if (!sp1) { free(buf); return -1; } - *sp1 = '\0'; - *out_method = el_strdup(buf); - char* path_start = sp1 + 1; - char* sp2 = strchr(path_start, ' '); - if (!sp2) { free(*out_method); *out_method = NULL; free(buf); return -1; } - *sp2 = '\0'; - *out_path = el_strdup(path_start); - char* hdr_end = strstr(sp2 + 1, "\r\n\r\n"); - /* Capture the raw header block (after the request line's CRLF, up to - * but not including the terminating \r\n\r\n) for callers that asked - * for it. The legacy 3-arg path passes NULL and skips this. */ - if (out_headers_block) { - char* hdr_start = strstr(sp2 + 1, "\r\n"); - if (hdr_start && hdr_start < hdr_end) { - hdr_start += 2; - size_t hb_len = (size_t)(hdr_end - hdr_start); - char* hb = malloc(hb_len + 1); - if (hb) { - memcpy(hb, hdr_start, hb_len); - hb[hb_len] = '\0'; - *out_headers_block = hb; - } - } else { - *out_headers_block = el_strdup(""); - } - } - /* Find Content-Length */ - long content_length = 0; - char* hp = sp2 + 1; - while (hp < hdr_end) { - char* line_end = strstr(hp, "\r\n"); - /* line_end == hdr_end means we're on the LAST header line — its - * trailing \r\n is the same \r\n that begins the \r\n\r\n header - * terminator. Process this line; only stop when line_end is past - * hdr_end (which means the parser walked off the end of the - * header block). The previous condition (line_end >= hdr_end) - * silently dropped any Content-Length that appeared as the last - * header — exactly what real curl/clients tend to emit. */ - if (!line_end || line_end > hdr_end) break; - if (strncasecmp(hp, "Content-Length:", 15) == 0) { - content_length = strtol(hp + 15, NULL, 10); - if (content_length < 0) content_length = 0; - if (content_length > 64 * 1024 * 1024) content_length = 64 * 1024 * 1024; - } - hp = line_end + 2; - } - /* Body: any bytes already read past hdr_end, plus more recv */ - char* body_start = hdr_end + 4; - size_t body_have = (buf + len) - body_start; - char* body = malloc((size_t)content_length + 1); - if (!body) { free(*out_method); free(*out_path); *out_method=NULL; *out_path=NULL; free(buf); return -1; } - if ((long)body_have > content_length) body_have = (size_t)content_length; - if (body_have > 0) memcpy(body, body_start, body_have); - while ((long)body_have < content_length) { - ssize_t n = recv(fd, body + body_have, (size_t)content_length - body_have, 0); - if (n <= 0) break; - body_have += (size_t)n; - } - body[body_have] = '\0'; - *out_body = body; - free(buf); - return 0; -} - -/* Reason phrase for common HTTP statuses. Falls back to "Status" for the - * long tail — clients only care about the numeric code. */ -static const char* http_reason_phrase(int status) { - switch (status) { - case 200: return "OK"; - case 201: return "Created"; - case 202: return "Accepted"; - case 204: return "No Content"; - case 301: return "Moved Permanently"; - case 302: return "Found"; - case 303: return "See Other"; - case 304: return "Not Modified"; - case 307: return "Temporary Redirect"; - case 308: return "Permanent Redirect"; - case 400: return "Bad Request"; - case 401: return "Unauthorized"; - case 403: return "Forbidden"; - case 404: return "Not Found"; - case 405: return "Method Not Allowed"; - case 409: return "Conflict"; - case 410: return "Gone"; - case 422: return "Unprocessable Entity"; - case 429: return "Too Many Requests"; - case 500: return "Internal Server Error"; - case 501: return "Not Implemented"; - case 502: return "Bad Gateway"; - case 503: return "Service Unavailable"; - case 504: return "Gateway Timeout"; - default: return "Status"; - } -} - -/* Best-effort send with retry on partial writes. */ -static int http_send_all(int fd, const char* p, size_t left) { - while (left > 0) { - ssize_t w = send(fd, p, left, 0); - if (w <= 0) return -1; - p += w; left -= (size_t)w; - } - return 0; -} - -/* Discriminator that http_response() embeds at the start of its envelope. - * A handler returning a string starting with this exact prefix is treated - * as a structured response; anything else is treated as a raw body. */ -#define EL_HTTP_RESPONSE_TAG "{\"el_http_response\":1" - -/* Keys that conflict with runtime-managed headers are silently dropped to - * avoid double-emission — the runtime always emits its own Content-Length - * and Connection: close. Content-Type from the envelope IS allowed and - * overrides auto-detection. */ -static int http_header_is_managed(const char* k) { - return strcasecmp(k, "Content-Length") == 0 - || strcasecmp(k, "Connection") == 0; -} - -/* Walk an ElMap of header pairs and emit each as `K: V\r\n` into JsonBuf b. - * Sets *out_saw_content_type to 1 if the map contained an explicit - * Content-Type so the caller can skip auto-detection. */ -static void http_emit_headers_from_map(JsonBuf* b, el_val_t headers_map, - int* out_saw_content_type) { - *out_saw_content_type = 0; - if (headers_map == 0) return; - ElMap* m = (ElMap*)(uintptr_t)headers_map; - if (!m || m->hdr.magic != EL_MAGIC_MAP) return; - for (int64_t i = 0; i < m->count; i++) { - const char* k = EL_CSTR(m->keys[i]); - const char* v = EL_CSTR(m->values[i]); - if (!k || !v) continue; - if (http_header_is_managed(k)) continue; - if (strcasecmp(k, "Content-Type") == 0) *out_saw_content_type = 1; - jb_puts(b, k); - jb_puts(b, ": "); - jb_puts(b, v); - jb_puts(b, "\r\n"); - } -} - -/* Parse the envelope produced by http_response(). On success returns 1 and - * populates *out_status, *out_headers_map (an ElMap el_val_t — caller must - * el_release), and *out_body (allocated). On failure returns 0. - * - * Implementation: feeds the entire envelope through the recursive-descent - * JSON parser (which builds proper ElMap/ElList values), then pulls the - * three top-level fields by name. Avoids re-stringifying the headers map - * since json_stringify() does not support nested objects. */ -static int http_parse_envelope(const char* s, int* out_status, - el_val_t* out_headers_map, char** out_body, - el_val_t* out_parsed_root) { - if (!s) return 0; - if (strncmp(s, EL_HTTP_RESPONSE_TAG, - sizeof(EL_HTTP_RESPONSE_TAG) - 1) != 0) return 0; - - el_val_t parsed = json_parse(EL_STR(s)); - if (parsed == EL_NULL) return 0; - - int status = 200; - el_val_t hmap = 0; - char* body = NULL; - - el_val_t sv = el_map_get(parsed, EL_STR("status")); - if (sv != 0) { - /* status comes back as an integer — el_val_t holds it directly. */ - long sc = (long)sv; - if (sc >= 100 && sc <= 599) status = (int)sc; - } - - el_val_t hv = el_map_get(parsed, EL_STR("headers")); - if (hv != 0) { - ElMap* hm = (ElMap*)(uintptr_t)hv; - if (hm && hm->hdr.magic == EL_MAGIC_MAP) hmap = hv; - } - - el_val_t bv = el_map_get(parsed, EL_STR("body")); - if (bv != 0) { - const char* bs = EL_CSTR(bv); - if (bs) body = el_strdup(bs); - } - if (!body) body = el_strdup(""); - - *out_status = status; - *out_headers_map = hmap; - *out_body = body; - *out_parsed_root = parsed; /* caller releases to free hmap + entries */ - return 1; -} - -/* Lightweight `__status__` envelope: if the body's first key is `__status__` - * and its value is a numeric literal, lift the status to the HTTP layer and - * strip the marker from the body before sending. This is the common case for - * El handlers that want to return 4xx/5xx without going through - * http_response() — they just prepend `{"__status__":,...}` to the JSON - * they were already returning. - * - * We deliberately recognise ONLY the first-key form so the contract is cheap - * to detect and unambiguous: `{"__status__":401,"error":"unauthorized"}` is - * an envelope, but `{"error":"...","__status__":401}` is not. Product code - * controls placement. - * - * On success returns 1 with *out_status set and *out_body_alloc populated - * with a freshly malloc'd body (caller frees). On failure returns 0 and - * leaves outputs untouched. */ -static int http_parse_status_envelope(const char* s, int* out_status, - char** out_body_alloc) { - if (!s) return 0; - const char* p = s; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '{') return 0; - p++; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - static const char marker[] = "\"__status__\""; - size_t mlen = sizeof(marker) - 1; - if (strncmp(p, marker, mlen) != 0) return 0; - p += mlen; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != ':') return 0; - p++; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p < '0' || *p > '9') return 0; /* non-numeric -> not an envelope */ - int status = 0; - while (*p >= '0' && *p <= '9') { - status = status * 10 + (*p - '0'); - p++; - } - if (status < 100 || status > 599) return 0; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - /* Two trailing shapes accepted: - * ,"k":v,...} -> body becomes {"k":v,...} - * } -> body becomes {} - * Anything else (e.g. `:` re-appearing, garbage) drops the envelope so - * we don't strip what we shouldn't. */ - if (*p == '}') { - *out_status = status; - *out_body_alloc = el_strdup("{}"); - return 1; - } - if (*p != ',') return 0; - p++; /* skip the comma; the rest of the object follows */ - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - /* Build the trimmed body: '{' + remainder. */ - size_t rest_len = strlen(p); - char* out = (char*)malloc(rest_len + 2); - if (!out) return 0; - out[0] = '{'; - memcpy(out + 1, p, rest_len); - out[rest_len + 1] = '\0'; - *out_status = status; - *out_body_alloc = out; - return 1; -} - -/* Send a fully-built HTTP response. If `body` starts with the envelope tag, - * unpack status/headers/body. Otherwise emit the historical 200-OK with - * auto-detected Content-Type. */ -/* Thread-local flag: if 1, http_send_response writes status + headers but - * NO body (HEAD method behaviour). Set by http_worker before calling - * http_send_response, cleared after. */ -static __thread int _tl_http_head_only = 0; - -static void http_send_response(int fd, const char* body) { - if (!body) body = ""; - - int status = 200; - el_val_t env_headers_map = 0; - char* env_body = NULL; - el_val_t env_parsed_root = 0; - int is_envelope = http_parse_envelope(body, &status, - &env_headers_map, &env_body, - &env_parsed_root); - - /* If the rich http_response() envelope didn't claim this body, try the - * lightweight `__status__` form. This second envelope is malloc-backed so - * we route it through env_body and let the existing cleanup path free it - * — same lifetime contract, no special case at the bottom of the - * function. */ - if (!is_envelope) { - char* trimmed = NULL; - if (http_parse_status_envelope(body, &status, &trimmed)) { - env_body = trimmed; - is_envelope = 1; - } - } - - const char* eff_body = is_envelope ? env_body : body; - /* Use the real byte count from fs_read if available (handles binary files - * with embedded null bytes — PNG, WOFF2, etc.). Fall back to strlen for - * normal text/JSON responses where _tl_fs_read_len is 0. */ - size_t blen = (_tl_fs_read_len > 0) ? _tl_fs_read_len : strlen(eff_body); - _tl_fs_read_len = 0; /* consume — one-shot per response */ - int head_only = _tl_http_head_only; - - JsonBuf hdrs; jb_init(&hdrs); - int saw_content_type = 0; - if (is_envelope) { - http_emit_headers_from_map(&hdrs, env_headers_map, - &saw_content_type); - } - if (!saw_content_type) { - jb_puts(&hdrs, "Content-Type: "); - jb_puts(&hdrs, http_detect_content_type(eff_body)); - jb_puts(&hdrs, "\r\n"); - } - - char status_line[64]; - int sl = snprintf(status_line, sizeof(status_line), - "HTTP/1.1 %d %s\r\n", - status, http_reason_phrase(status)); - if (sl < 0) { - if (env_parsed_root) el_release(env_parsed_root); - free(env_body); free(hdrs.buf); return; - } - - char tail[128]; - int tl = snprintf(tail, sizeof(tail), - "Content-Length: %zu\r\n" - "Connection: close\r\n" - "\r\n", blen); - if (tl < 0) { - if (env_parsed_root) el_release(env_parsed_root); - free(env_body); free(hdrs.buf); return; - } - - if (http_send_all(fd, status_line, (size_t)sl) == 0 - && http_send_all(fd, hdrs.buf, hdrs.len) == 0 - && http_send_all(fd, tail, (size_t)tl) == 0 - && (head_only - /* HEAD requests echo headers + Content-Length but no body. */ - ? 1 - : http_send_all(fd, eff_body, blen) == 0)) { - /* sent successfully */ - } - - if (env_parsed_root) el_release(env_parsed_root); - free(env_body); - free(hdrs.buf); -} - -typedef struct { - int fd; -} HttpWorkerArg; - -static void* http_worker(void* arg) { - HttpWorkerArg* a = (HttpWorkerArg*)arg; - int fd = a->fd; - free(a); - char *method = NULL, *path = NULL, *body = NULL; - if (http_read_request(fd, &method, &path, &body, NULL) == 0) { - http_handler_fn h = http_lookup_active(); - char* response = NULL; - /* HEAD: dispatch as GET so existing handlers respond with the same - * body, but flag the response writer to emit headers only. RFC 9110 - * requires HEAD to mirror GET headers + Content-Length without body. */ - int head_only = (method && strcmp(method, "HEAD") == 0); - const char* dispatch_method = head_only ? "GET" : method; - el_request_start(); /* begin per-request arena */ - if (h) { - el_val_t r = h(EL_STR(dispatch_method), EL_STR(path), EL_STR(body)); - const char* rs = EL_CSTR(r); - /* Copy response out BEFORE arena teardown. - * For binary files, _tl_fs_read_len holds the real byte count — - * use memcpy instead of strdup so null bytes are preserved. */ - size_t rlen = _tl_fs_read_len > 0 ? _tl_fs_read_len : (rs ? strlen(rs) : 0); - response = malloc(rlen + 1); - if (response && rs) { memcpy(response, rs, rlen); response[rlen] = '\0'; } - else if (response) { response[0] = '\0'; } - } else { - response = el_strdup_persist("el-runtime: no http handler registered"); - } - el_request_end(); /* free all intermediate strings */ - _tl_http_head_only = head_only; - http_send_response(fd, response); - _tl_http_head_only = 0; - free(response); - } - free(method); free(path); free(body); - close(fd); - /* release a slot */ - pthread_mutex_lock(&_http_conn_mu); - _http_conn_active--; - pthread_cond_signal(&_http_conn_cv); - pthread_mutex_unlock(&_http_conn_mu); - return NULL; -} - -void http_serve(el_val_t port, el_val_t handler) { - /* If `handler` looks like a string name, register it as the active handler. */ - const char* hname = EL_CSTR(handler); - if (hname && looks_like_string(handler)) { - http_set_handler(handler); - } - int p = (int)port; - if (p <= 0 || p > 65535) { fprintf(stderr, "http_serve: invalid port %d\n", p); return; } - /* Dual-stack: AF_INET6 with IPV6_V6ONLY=0 accepts both IPv4 and IPv6. - * This makes `localhost` work in browsers that resolve it to ::1 first. */ - int sock = socket(AF_INET6, SOCK_STREAM, 0); - if (sock < 0) { perror("socket"); return; } - int yes = 1; int no = 0; - setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes)); - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &no, sizeof(no)); - struct sockaddr_in6 addr; - memset(&addr, 0, sizeof(addr)); - addr.sin6_family = AF_INET6; - addr.sin6_addr = in6addr_any; - addr.sin6_port = htons((uint16_t)p); - if (bind(sock, (struct sockaddr*)&addr, sizeof(addr)) < 0) { - perror("bind"); close(sock); return; - } - if (listen(sock, 64) < 0) { perror("listen"); close(sock); return; } - fprintf(stderr, "[http] listening on [::]:%d (dual-stack)\n", p); - while (1) { - struct sockaddr_in6 cli; - socklen_t clen = sizeof(cli); - int cfd = accept(sock, (struct sockaddr*)&cli, &clen); - if (cfd < 0) { - if (errno == EINTR) continue; - perror("accept"); break; - } - pthread_mutex_lock(&_http_conn_mu); - while (_http_conn_active >= HTTP_MAX_CONNS) { - pthread_cond_wait(&_http_conn_cv, &_http_conn_mu); - } - _http_conn_active++; - pthread_mutex_unlock(&_http_conn_mu); - HttpWorkerArg* arg = malloc(sizeof(HttpWorkerArg)); - if (!arg) { close(cfd); continue; } - arg->fd = cfd; - pthread_t tid; - if (pthread_create(&tid, NULL, http_worker, arg) != 0) { - close(cfd); free(arg); - pthread_mutex_lock(&_http_conn_mu); - _http_conn_active--; - pthread_cond_signal(&_http_conn_cv); - pthread_mutex_unlock(&_http_conn_mu); - continue; - } - pthread_detach(tid); - } - close(sock); -} - -/* ── HTTP server v2 — request headers + structured response ──────────────── */ -/* - * v2 widens the handler signature from - * (method, path, body) -> body_string - * to - * (method, path, headers_map, body) -> body_string_or_envelope - * - * The response envelope is detected uniformly inside http_send_response — so - * 4-arg handlers can return either a plain body or http_response(...). The - * 3-arg path stays untouched in spirit (its handlers still build plain - * bodies; the envelope tag, being `{"el_http_response":1`, will never - * collide with normal JSON the legacy server.el routes return). - * - * Registry is parallel to the 3-arg handler registry: separate name table, - * separate active-handler slot, separate dlsym fallback. Mixing v1 and v2 - * handlers in the same process is fine — they don't share the active slot. */ - -typedef el_val_t (*http_handler4_fn)(el_val_t method, el_val_t path, - el_val_t headers_map, el_val_t body); - -typedef struct { - char* name; - http_handler4_fn fn; -} HttpHandler4Entry; - -static HttpHandler4Entry _http_handlers4[32]; -static size_t _http_handler4_count = 0; -static char* _http_active_handler4 = NULL; - -void el_runtime_register_handler_v2(const char* name, http_handler4_fn fn); -void el_runtime_register_handler_v2(const char* name, http_handler4_fn fn) { - if (!name || !fn) return; - pthread_mutex_lock(&_http_handler_mu); - for (size_t i = 0; i < _http_handler4_count; i++) { - if (strcmp(_http_handlers4[i].name, name) == 0) { - _http_handlers4[i].fn = fn; - pthread_mutex_unlock(&_http_handler_mu); - return; - } - } - if (_http_handler4_count < - sizeof(_http_handlers4) / sizeof(_http_handlers4[0])) { - _http_handlers4[_http_handler4_count].name = el_strdup(name); - _http_handlers4[_http_handler4_count].fn = fn; - _http_handler4_count++; - } - pthread_mutex_unlock(&_http_handler_mu); -} - -void http_set_handler_v2(el_val_t name) { - const char* n = EL_CSTR(name); - pthread_mutex_lock(&_http_handler_mu); - free(_http_active_handler4); - _http_active_handler4 = el_strdup(n ? n : ""); - if (n && *n) { - int found = 0; - for (size_t i = 0; i < _http_handler4_count; i++) { - if (strcmp(_http_handlers4[i].name, n) == 0) { found = 1; break; } - } - if (!found) { - void* sym = dlsym(RTLD_DEFAULT, n); - if (sym && _http_handler4_count < - sizeof(_http_handlers4) / sizeof(_http_handlers4[0])) { - _http_handlers4[_http_handler4_count].name = el_strdup(n); - _http_handlers4[_http_handler4_count].fn = - (http_handler4_fn)sym; - _http_handler4_count++; - } - } - } - pthread_mutex_unlock(&_http_handler_mu); -} - -static http_handler4_fn http_lookup_active_v2(void) { - http_handler4_fn out = NULL; - pthread_mutex_lock(&_http_handler_mu); - if (_http_active_handler4) { - for (size_t i = 0; i < _http_handler4_count; i++) { - if (strcmp(_http_handlers4[i].name, - _http_active_handler4) == 0) { - out = _http_handlers4[i].fn; break; - } - } - } - pthread_mutex_unlock(&_http_handler_mu); - return out; -} - -/* Build an ElMap from the raw header block produced by http_read_request. - * Keys are lowercased (RFC 7230 — case-insensitive); values have leading - * whitespace trimmed. Repeated headers with the same name are joined with - * ", " in arrival order, matching standard library behaviour elsewhere. */ -static el_val_t http_build_headers_map(const char* hdr_block) { - el_val_t m = el_map_new(0); - if (!hdr_block || !*hdr_block) return m; - const char* p = hdr_block; - while (*p) { - const char* line_end = strstr(p, "\r\n"); - const char* end = line_end ? line_end : p + strlen(p); - const char* colon = NULL; - for (const char* c = p; c < end; c++) { - if (*c == ':') { colon = c; break; } - } - if (colon && colon > p) { - size_t klen = (size_t)(colon - p); - char* key = malloc(klen + 1); - if (key) { - for (size_t i = 0; i < klen; i++) { - unsigned char ch = (unsigned char)p[i]; - key[i] = (char)tolower(ch); - } - key[klen] = '\0'; - const char* vstart = colon + 1; - while (vstart < end && (*vstart == ' ' || *vstart == '\t')) vstart++; - size_t vlen = (size_t)(end - vstart); - /* Strip trailing OWS just in case. */ - while (vlen > 0 - && (vstart[vlen - 1] == ' ' - || vstart[vlen - 1] == '\t')) vlen--; - /* Coalesce repeats: if key already present, append ", value". */ - el_val_t existing = el_map_get(m, EL_STR(key)); - if (existing != 0 && looks_like_string(existing)) { - const char* old = EL_CSTR(existing); - size_t olen = strlen(old); - char* combined = malloc(olen + 2 + vlen + 1); - if (combined) { - memcpy(combined, old, olen); - memcpy(combined + olen, ", ", 2); - memcpy(combined + olen + 2, vstart, vlen); - combined[olen + 2 + vlen] = '\0'; - m = el_map_set(m, EL_STR(key), EL_STR(combined)); - } - free(key); - } else { - char* val = malloc(vlen + 1); - if (val) { - memcpy(val, vstart, vlen); - val[vlen] = '\0'; - m = el_map_set(m, EL_STR(key), EL_STR(val)); - } else { - free(key); - } - } - } - } - if (!line_end) break; - p = line_end + 2; - } - return m; -} - -static void* http_worker_v2(void* arg) { - HttpWorkerArg* a = (HttpWorkerArg*)arg; - int fd = a->fd; - free(a); - char *method = NULL, *path = NULL, *body = NULL, *hdr_block = NULL; - if (http_read_request(fd, &method, &path, &body, &hdr_block) == 0) { - http_handler4_fn h = http_lookup_active_v2(); - char* response = NULL; - int head_only = (method && strcmp(method, "HEAD") == 0); - const char* dispatch_method = head_only ? "GET" : method; - el_request_start(); /* begin per-request arena */ - if (h) { - el_val_t hmap = http_build_headers_map(hdr_block ? hdr_block : ""); - el_val_t r = h(EL_STR(dispatch_method), EL_STR(path), hmap, EL_STR(body)); - const char* rs = EL_CSTR(r); - size_t rlen = _tl_fs_read_len > 0 ? _tl_fs_read_len : (rs ? strlen(rs) : 0); - response = malloc(rlen + 1); - if (response && rs) { memcpy(response, rs, rlen); response[rlen] = '\0'; } - else if (response) { response[0] = '\0'; } - el_release(hmap); - } else { - response = el_strdup_persist( - "el-runtime: no v2 http handler registered " - "(call http_set_handler_v2)"); - } - el_request_end(); /* free all intermediate strings */ - _tl_http_head_only = head_only; - http_send_response(fd, response); - _tl_http_head_only = 0; - free(response); - } - free(method); free(path); free(body); free(hdr_block); - close(fd); - pthread_mutex_lock(&_http_conn_mu); - _http_conn_active--; - pthread_cond_signal(&_http_conn_cv); - pthread_mutex_unlock(&_http_conn_mu); - return NULL; -} - -void http_serve_v2(el_val_t port, el_val_t handler) { - const char* hname = EL_CSTR(handler); - if (hname && looks_like_string(handler)) { - http_set_handler_v2(handler); - } - int p = (int)port; - if (p <= 0 || p > 65535) { - fprintf(stderr, "http_serve_v2: invalid port %d\n", p); - return; - } - /* Dual-stack: same as http_serve - AF_INET6 + IPV6_V6ONLY=0. */ - int sock = socket(AF_INET6, SOCK_STREAM, 0); - if (sock < 0) { perror("socket"); return; } - int yes = 1; int no = 0; - setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes)); - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &no, sizeof(no)); - struct sockaddr_in6 addr; - memset(&addr, 0, sizeof(addr)); - addr.sin6_family = AF_INET6; - addr.sin6_addr = in6addr_any; - addr.sin6_port = htons((uint16_t)p); - if (bind(sock, (struct sockaddr*)&addr, sizeof(addr)) < 0) { - perror("bind"); close(sock); return; - } - if (listen(sock, 64) < 0) { perror("listen"); close(sock); return; } - fprintf(stderr, "[http v2] listening on [::]:%d (dual-stack)\n", p); - while (1) { - struct sockaddr_in6 cli; - socklen_t clen = sizeof(cli); - int cfd = accept(sock, (struct sockaddr*)&cli, &clen); - if (cfd < 0) { - if (errno == EINTR) continue; - perror("accept"); break; - } - pthread_mutex_lock(&_http_conn_mu); - while (_http_conn_active >= HTTP_MAX_CONNS) { - pthread_cond_wait(&_http_conn_cv, &_http_conn_mu); - } - _http_conn_active++; - pthread_mutex_unlock(&_http_conn_mu); - HttpWorkerArg* arg = malloc(sizeof(HttpWorkerArg)); - if (!arg) { close(cfd); continue; } - arg->fd = cfd; - pthread_t tid; - if (pthread_create(&tid, NULL, http_worker_v2, arg) != 0) { - close(cfd); free(arg); - pthread_mutex_lock(&_http_conn_mu); - _http_conn_active--; - pthread_cond_signal(&_http_conn_cv); - pthread_mutex_unlock(&_http_conn_mu); - continue; - } - pthread_detach(tid); - } - close(sock); -} - -/* Build the response envelope a 4-arg handler can return. We hand-write - * the JSON so the discriminator key always lands first — the runtime's - * http_parse_envelope() detects it via prefix match. headers_json must be - * either "" (empty), "{}" (empty object), or a well-formed JSON object - * literal; anything else will produce a malformed envelope and the runtime - * will treat the whole string as a plain body (no envelope detected). */ -el_val_t http_response(el_val_t status, el_val_t headers_json, el_val_t body) { - long sc = (long)status; - if (sc < 100 || sc > 599) sc = 200; - const char* hj = EL_CSTR(headers_json); - if (!hj || !*hj) hj = "{}"; - /* Light validation: must start with '{' and end with '}'. */ - size_t hlen = strlen(hj); - int hj_ok = (hlen >= 2 && hj[0] == '{' && hj[hlen - 1] == '}'); - if (!hj_ok) hj = "{}"; - const char* b = EL_CSTR(body); - if (!b) b = ""; - - JsonBuf out; jb_init(&out); - jb_puts(&out, EL_HTTP_RESPONSE_TAG); /* {"el_http_response":1 */ - jb_puts(&out, ",\"status\":"); - char num[32]; - snprintf(num, sizeof(num), "%ld", sc); - jb_puts(&out, num); - jb_puts(&out, ",\"headers\":"); - jb_puts(&out, hj); - jb_puts(&out, ",\"body\":"); - jb_emit_escaped(&out, b); - jb_putc(&out, '}'); - return el_wrap_str(out.buf); -} - -/* ── Filesystem ──────────────────────────────────────────────────────────── */ - -el_val_t fs_read(el_val_t pathv) { - const char* path = EL_CSTR(pathv); - _tl_fs_read_len = 0; - if (!path) return el_wrap_str(el_strdup("")); - FILE* f = fopen(path, "rb"); - if (!f) return el_wrap_str(el_strdup("")); - fseek(f, 0, SEEK_END); - long sz = ftell(f); - rewind(f); - if (sz < 0) { fclose(f); return el_wrap_str(el_strdup("")); } /* pipe/special file */ - char* buf = el_strbuf((size_t)sz); - size_t got = fread(buf, 1, (size_t)sz, f); - buf[got] = '\0'; - _tl_fs_read_len = got; /* store real byte count for binary-safe send */ - fclose(f); - return el_wrap_str(buf); -} - -el_val_t fs_write(el_val_t pathv, el_val_t contentv) { - const char* path = EL_CSTR(pathv); - const char* content = EL_CSTR(contentv); - if (!path || !content) return 0; - FILE* f = fopen(path, "wb"); - if (!f) return 0; - size_t n = strlen(content); - size_t written = fwrite(content, 1, n, f); - fclose(f); - return written == n ? 1 : 0; -} - -/* fs_write_bytes — explicit-length binary write. Bypasses strlen so embedded - * NULs survive. Caller must know the byte count (e.g. from base64_decode, - * or the fixed 32-byte sha256_bytes/hmac_sha256_bytes outputs). - * - * If `length` is negative, treats as failure. If `length` is 0, creates an - * empty file (still useful as a "touch with content" primitive). */ -el_val_t fs_write_bytes(el_val_t pathv, el_val_t bytesv, el_val_t lengthv) { - const char* path = EL_CSTR(pathv); - const char* bytes = EL_CSTR(bytesv); - int64_t n = (int64_t)lengthv; - if (!path || !bytes) return 0; - if (n < 0) return 0; - FILE* f = fopen(path, "wb"); - if (!f) return 0; - size_t written = (n > 0) ? fwrite(bytes, 1, (size_t)n, f) : 0; - int flush_ok = (fflush(f) == 0); - int close_ok = (fclose(f) == 0); - if (!flush_ok || !close_ok || written != (size_t)n) { - remove(path); - return 0; - } - return 1; -} - -// exec_command — run a shell command, return exit code (0 = success). -// Used by elb and other El tooling to invoke subprocesses. -el_val_t exec_command(el_val_t cmdv) { - const char* cmd = EL_CSTR(cmdv); - if (!cmd) return (el_val_t)(int64_t)-1; - int ret = system(cmd); - return (el_val_t)(int64_t)ret; -} - -// exec_capture — run a shell command, capture stdout, return as String. -// Returns "" on failure. -el_val_t exec_capture(el_val_t cmdv) { - const char* cmd = EL_CSTR(cmdv); - if (!cmd) return el_wrap_str(el_strdup("")); - FILE* f = popen(cmd, "r"); - if (!f) return el_wrap_str(el_strdup("")); - JsonBuf b; jb_init(&b); - char buf[4096]; - while (fgets(buf, sizeof(buf), f)) jb_puts(&b, buf); - pclose(f); - return el_wrap_str(b.buf); -} - -// exec — run a shell command via /bin/sh, capture stdout, return as String. -// Times out after 30 seconds. Returns "" on any error. -// El name: exec(cmd) -> String -el_val_t exec(el_val_t cmdv) { - const char* cmd = EL_CSTR(cmdv); - if (!cmd || !*cmd) return el_wrap_str(el_strdup("")); - /* Build a time-limited command: wrap with timeout(1) if available, - * otherwise rely on the 30s read loop guard below. We use the simple - * popen approach with a deadline measured by wall clock so the caller - * is never blocked indefinitely. */ - FILE* f = popen(cmd, "r"); - if (!f) return el_wrap_str(el_strdup("")); - JsonBuf b; jb_init(&b); - char buf[4096]; - /* 30-second wall-clock deadline */ - time_t deadline = time(NULL) + 30; - while (time(NULL) < deadline) { - if (fgets(buf, sizeof(buf), f) == NULL) break; - jb_puts(&b, buf); - } - pclose(f); - return el_wrap_str(b.buf); -} - -// exec_bg — run a shell command in background, return PID as String. -// The child process runs independently; the caller is not blocked. -// Returns "" on fork failure. -// El name: exec_bg(cmd) -> String -el_val_t exec_bg(el_val_t cmdv) { - const char* cmd = EL_CSTR(cmdv); - if (!cmd || !*cmd) return el_wrap_str(el_strdup("")); - pid_t pid = fork(); - if (pid < 0) { - /* fork failed */ - return el_wrap_str(el_strdup("")); - } - if (pid == 0) { - /* child: detach from parent's stdio, exec via shell */ - setsid(); - int devnull = open("/dev/null", O_RDWR); - if (devnull >= 0) { - dup2(devnull, STDIN_FILENO); - dup2(devnull, STDOUT_FILENO); - dup2(devnull, STDERR_FILENO); - close(devnull); - } - execl("/bin/sh", "sh", "-c", cmd, (char*)NULL); - _exit(127); - } - /* parent: convert pid to string and return immediately */ - char pidbuf[32]; - snprintf(pidbuf, sizeof(pidbuf), "%d", (int)pid); - return el_wrap_str(el_strdup(pidbuf)); -} - -el_val_t fs_list(el_val_t pathv) { - const char* path = EL_CSTR(pathv); - el_val_t lst = el_list_empty(); - if (!path) return lst; - DIR* d = opendir(path); - if (!d) return lst; - struct dirent* e; - while ((e = readdir(d)) != NULL) { - if (strcmp(e->d_name, ".") == 0 || strcmp(e->d_name, "..") == 0) continue; - lst = el_list_append(lst, el_wrap_str(el_strdup(e->d_name))); - } - closedir(d); - return lst; -} - -/* fs_exists — true iff stat(path) succeeds. Symlinks are followed. */ -el_val_t fs_exists(el_val_t pathv) { - const char* path = EL_CSTR(pathv); - if (!path || !*path) return 0; - struct stat st; - return (el_val_t)(stat(path, &st) == 0 ? 1 : 0); -} - -/* fs_mkdir — create directory at path with mode 0755, mkdir -p semantics. - * Returns 1 if path exists or was created (incl. all parents); 0 on failure. - * Walks the path component-by-component so missing intermediate dirs are - * also created. An existing leaf is not an error. */ -el_val_t fs_mkdir(el_val_t pathv) { - const char* path = EL_CSTR(pathv); - if (!path || !*path) return 0; - size_t n = strlen(path); - char* buf = malloc(n + 1); - if (!buf) return 0; - memcpy(buf, path, n + 1); - /* Walk components; create each prefix in turn. */ - for (size_t i = 1; i <= n; i++) { - if (buf[i] == '/' || buf[i] == '\0') { - char saved = buf[i]; - buf[i] = '\0'; - if (buf[0] != '\0') { - if (mkdir(buf, 0755) != 0 && errno != EEXIST) { - /* Tolerate the case where this prefix exists as a non-dir - * only when stat says it's a directory. */ - struct stat st; - if (stat(buf, &st) != 0 || !S_ISDIR(st.st_mode)) { - free(buf); - return 0; - } - } - } - buf[i] = saved; - } - } - free(buf); - return 1; -} - -/* ── URL encoding ─────────────────────────────────────────────────────────── */ - -/* RFC 3986 percent-encoding for URL components (form bodies, query strings). - * Unreserved set: A-Z a-z 0-9 - _ . ~ — passed through verbatim. - * Everything else (including space) becomes %XX hex. */ -el_val_t url_encode(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - static const char hex[] = "0123456789ABCDEF"; - size_t n = strlen(s); - char* out = el_strbuf(n * 3); - size_t o = 0; - for (size_t i = 0; i < n; i++) { - unsigned char c = (unsigned char)s[i]; - if ((c >= 'A' && c <= 'Z') || - (c >= 'a' && c <= 'z') || - (c >= '0' && c <= '9') || - c == '-' || c == '_' || c == '.' || c == '~') { - out[o++] = (char)c; - } else { - out[o++] = '%'; - out[o++] = hex[(c >> 4) & 0xF]; - out[o++] = hex[c & 0xF]; - } - } - out[o] = '\0'; - return el_wrap_str(out); -} - -/* Decode percent-encoded URL component. '+' becomes space (form-encoded); - * malformed %-escapes are emitted verbatim. */ -el_val_t url_decode(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - size_t n = strlen(s); - char* out = el_strbuf(n); - size_t o = 0; - for (size_t i = 0; i < n; i++) { - char c = s[i]; - if (c == '+') { - out[o++] = ' '; - } else if (c == '%' && i + 2 < n) { - char h1 = s[i + 1], h2 = s[i + 2]; - int v1 = (h1 >= '0' && h1 <= '9') ? h1 - '0' - : (h1 >= 'a' && h1 <= 'f') ? h1 - 'a' + 10 - : (h1 >= 'A' && h1 <= 'F') ? h1 - 'A' + 10 : -1; - int v2 = (h2 >= '0' && h2 <= '9') ? h2 - '0' - : (h2 >= 'a' && h2 <= 'f') ? h2 - 'a' + 10 - : (h2 >= 'A' && h2 <= 'F') ? h2 - 'A' + 10 : -1; - if (v1 >= 0 && v2 >= 0) { - out[o++] = (char)((v1 << 4) | v2); - i += 2; - } else { - out[o++] = c; - } - } else { - out[o++] = c; - } - } - out[o] = '\0'; - return el_wrap_str(out); -} - -/* ── HTML allowlist sanitizer ──────────────────────────────────────────────── - * el_html_sanitize(input, allowlist_json) - * - * Strict allowlist HTML cleaner. Replaces the older denylist patterns - * (str_replace cascades that wrapped dangerous tags in HTML comments and - * renamed `on*` attributes). The denylist approach is fragile: comment- - * wrapping can be re-broken by a literal `-->` inside an attacker-supplied - * attribute value, and every new attack vector requires a code change. - * - * Design: - * - Single-pass byte-level state machine. - * - Tag and attribute names are matched case-insensitively against the - * allowlist. Unknown tags are dropped entirely (the open and close - * markers are stripped; their inner text content survives, escaped). - * - A small set of "dangerous container" tags (script, style, iframe, - * object, embed, form, plus a few rarer ones) drop themselves AND - * their full subtree — text between `` is - * CDATA-like and must not be re-emitted as escaped text either. - * - Comments (), doctype (), CDATA (), - * and processing instructions () are dropped entirely. - * - Text content outside dropped subtrees is HTML-escaped (&, <, >, ", '). - * - Attribute values are unquoted/dequoted, then re-emitted with double - * quotes around the cleanly-escaped value. - * - For `` and any `src` attribute, the URL scheme is validated: - * only http:, https:, mailto:, fragment-only `#anchor`, or relative - * paths are allowed. Anything else (javascript:, data:, vbscript:, - * about:, file:, etc.) drops the attribute. - * - Self-closing void tags (br, hr, img, etc.) emit without a close tag. - * - Malformed input (unclosed tag at EOF, bad attribute syntax) drops - * the pending tag and continues. Pre-encoded entities (<, &, - * etc.) are passed through verbatim — the browser will decode them - * safely on render. - * - * Allowlist format (JSON string): - * {"p":[],"a":["href","title"],"strong":[],...} - * - Key = lowercase tag name. - * - Value = JSON array of allowed attribute names (lowercase). - * - Empty array means tag allowed but no attributes survive. - * - * Output is a freshly-allocated arena-tracked el_val_t string. */ - -/* Internal byte buffer with realloc-doubling. Used during sanitization; - * the final result is copied into an arena-tracked el_strbuf so the caller - * sees standard runtime memory semantics. */ -typedef struct { - char* data; - size_t len; - size_t cap; -} html_buf_t; - -static void html_buf_init(html_buf_t* b) { - b->cap = 256; - b->data = malloc(b->cap); - if (!b->data) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - b->len = 0; -} - -static void html_buf_grow(html_buf_t* b, size_t need) { - if (b->len + need + 1 <= b->cap) return; - size_t nc = b->cap; - while (b->len + need + 1 > nc) nc *= 2; - char* nd = realloc(b->data, nc); - if (!nd) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - b->data = nd; - b->cap = nc; -} - -static void html_buf_putc(html_buf_t* b, char c) { - html_buf_grow(b, 1); - b->data[b->len++] = c; -} - -static void html_buf_puts(html_buf_t* b, const char* s) { - if (!s) return; - size_t n = strlen(s); - html_buf_grow(b, n); - memcpy(b->data + b->len, s, n); - b->len += n; -} - -static void html_buf_free(html_buf_t* b) { - free(b->data); - b->data = NULL; - b->len = b->cap = 0; -} - -/* ASCII tolower, locale-independent. */ -static int html_tolower(int c) { - return (c >= 'A' && c <= 'Z') ? c + 32 : c; -} - -/* Case-insensitive ASCII compare of [a, a+n) against c-string `s`. - * Returns 1 iff lengths match and bytes are equal under tolower. */ -static int html_ieq_n(const char* a, size_t n, const char* s) { - if (!a || !s) return 0; - if (strlen(s) != n) return 0; - for (size_t i = 0; i < n; i++) { - if (html_tolower((unsigned char)a[i]) != html_tolower((unsigned char)s[i])) return 0; - } - return 1; -} - -/* Case-insensitive ASCII compare of two byte slices. */ -static int html_iemem(const char* a, const char* b, size_t n) { - for (size_t i = 0; i < n; i++) { - if (html_tolower((unsigned char)a[i]) != html_tolower((unsigned char)b[i])) return 0; - } - return 1; -} - -/* Walk a JSON allowlist object and find the value (an array) for a given - * tag key, comparing case-insensitively. On hit returns a pointer to the - * opening `[` of the array and writes the byte length of the array span - * (including the brackets) to *out_len. On miss returns NULL. - * - * The parser is intentionally tiny: it does not handle escapes inside - * keys (allowlist authors do not need them), and it relies on balanced - * brackets/quotes within the value array. */ -static const char* html_allowlist_find(const char* allow, const char* tag, - size_t tag_len, size_t* out_len) { - if (!allow) return NULL; - const char* p = allow; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '{') return NULL; - p++; - while (*p) { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - if (*p == '}' || *p == 0) return NULL; - if (*p != '"') return NULL; - p++; - const char* k = p; - while (*p && *p != '"') p++; - if (*p != '"') return NULL; - size_t klen = (size_t)(p - k); - p++; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != ':') return NULL; - p++; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '[') return NULL; - const char* arr_start = p; - int depth = 0; - int in_str = 0; - while (*p) { - char c = *p; - if (in_str) { - if (c == '\\' && p[1]) { p += 2; continue; } - if (c == '"') in_str = 0; - } else { - if (c == '"') in_str = 1; - else if (c == '[') depth++; - else if (c == ']') { depth--; if (depth == 0) { p++; break; } } - } - p++; - } - size_t alen = (size_t)(p - arr_start); - int match = (klen == tag_len) && html_iemem(k, tag, klen); - if (match) { - if (out_len) *out_len = alen; - return arr_start; - } - } - return NULL; -} - -/* Returns 1 iff `attr` (length attr_len) appears as a string element - * in the JSON array slice [arr, arr+arr_len). Comparison is case- - * insensitive. */ -static int html_attr_in_array(const char* arr, size_t arr_len, - const char* attr, size_t attr_len) { - if (!arr || arr_len < 2) return 0; - const char* p = arr + 1; - const char* end = arr + arr_len - 1; - while (p < end) { - while (p < end && (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',')) p++; - if (p >= end) return 0; - if (*p != '"') return 0; - p++; - const char* s = p; - while (p < end && *p != '"') { - if (*p == '\\' && p + 1 < end) p++; - p++; - } - if (p >= end) return 0; - size_t slen = (size_t)(p - s); - p++; - if (slen == attr_len && html_iemem(s, attr, slen)) return 1; - } - return 0; -} - -/* Hard-coded set of tags whose content is ALSO dropped (entire subtree). */ -static int html_is_dangerous_container(const char* tag, size_t tag_len) { - static const char* names[] = { - "script", "style", "iframe", "object", "embed", "form", - "noscript", "noembed", "template", "svg", "math", "frame", - "frameset", "applet", "audio", "video", "source", "track", - NULL - }; - for (int i = 0; names[i]; i++) { - if (html_ieq_n(tag, tag_len, names[i])) return 1; - } - return 0; -} - -/* HTML void elements — emit without a close tag. */ -static int html_is_void(const char* tag, size_t tag_len) { - static const char* names[] = { - "area", "base", "br", "col", "embed", "hr", "img", "input", - "link", "meta", "param", "source", "track", "wbr", - NULL - }; - for (int i = 0; names[i]; i++) { - if (html_ieq_n(tag, tag_len, names[i])) return 1; - } - return 0; -} - -/* Append a single byte HTML-escaped into the output buffer. */ -static void html_escape_byte(html_buf_t* out, unsigned char c) { - switch (c) { - case '<': html_buf_puts(out, "<"); break; - case '>': html_buf_puts(out, ">"); break; - case '"': html_buf_puts(out, """); break; - case '\'': html_buf_puts(out, "'"); break; - default: html_buf_putc(out, (char)c); break; - } -} - -/* Validate a URL value against the allowlist of safe schemes for hrefs. - * Returns 1 iff the URL is safe to emit. Acceptable forms: - * - http:// or https:// (case-insensitive) - * - mailto: - * - fragment-only `#anchor` - * - relative path that does not contain a colon before the first - * slash/?/# (so `foo/bar`, `/foo`, `?x=1` are OK; `javascript:x` is - * not — its colon precedes any path/hash/query separator). - * - * URL leading whitespace and embedded ASCII control bytes (TAB, LF, CR) - * are stripped before the scheme test, mirroring how browsers normalise - * URLs (these bytes are otherwise a known XSS bypass: `java\tscript:`). */ -static int html_url_is_safe(const char* url, size_t len) { - if (!url || len == 0) return 1; /* empty href is harmless */ - size_t i = 0; - while (i < len) { - unsigned char c = (unsigned char)url[i]; - if (c == ' ' || c == '\t' || c == '\n' || c == '\r' || c == 0x0B || c == 0x0C) { - i++; continue; - } - break; - } - if (i >= len) return 1; /* whitespace only */ - if (url[i] == '#') return 1; /* fragment only */ - if (url[i] == '/' || url[i] == '?') return 1; /* relative */ - /* Find the first scheme-terminating character. */ - size_t scheme_end = (size_t)-1; - for (size_t j = i; j < len; j++) { - char c = url[j]; - if (c == ':') { scheme_end = j; break; } - if (c == '/' || c == '?' || c == '#') break; - } - if (scheme_end == (size_t)-1) return 1; /* no colon → relative path */ - /* Lowercase the scheme, stripping embedded control bytes. */ - char scheme[32]; - size_t sl = 0; - for (size_t j = i; j < scheme_end && sl < sizeof(scheme) - 1; j++) { - unsigned char c = (unsigned char)url[j]; - if (c == '\t' || c == '\n' || c == '\r' || c == 0x0B || c == 0x0C) continue; - scheme[sl++] = (char)html_tolower(c); - } - scheme[sl] = '\0'; - if (strcmp(scheme, "http") == 0) return 1; - if (strcmp(scheme, "https") == 0) return 1; - if (strcmp(scheme, "mailto") == 0) return 1; - return 0; -} - -el_val_t el_html_sanitize(el_val_t input_v, el_val_t allowlist_v) { - const char* input = EL_CSTR(input_v); - const char* allow = EL_CSTR(allowlist_v); - if (!input) return el_wrap_str(el_strdup("")); - if (!allow) allow = "{}"; - size_t in_len = strlen(input); - - html_buf_t out; - html_buf_init(&out); - - size_t i = 0; - while (i < in_len) { - unsigned char c = (unsigned char)input[i]; - if (c != '<') { - /* Plain text — escape and emit. We pass `&` through verbatim - * to preserve pre-encoded entities (`<`, `&`, `&#x...;`) - * which the browser will decode safely. */ - if (c == '&') html_buf_putc(&out, '&'); - else html_escape_byte(&out, c); - i++; - continue; - } - /* `<` — try to parse a tag. */ - if (i + 1 >= in_len) { - html_buf_puts(&out, "<"); - i++; - continue; - } - /* Comments, doctype, CDATA, processing instructions — drop entirely. */ - if (input[i + 1] == '!') { - if (i + 3 < in_len && input[i + 2] == '-' && input[i + 3] == '-') { - size_t j = i + 4; - while (j + 2 < in_len && !(input[j] == '-' && input[j + 1] == '-' && input[j + 2] == '>')) j++; - if (j + 2 < in_len) i = j + 3; - else i = in_len; - continue; - } - size_t j = i + 2; - while (j < in_len && input[j] != '>') j++; - i = (j < in_len) ? j + 1 : in_len; - continue; - } - if (input[i + 1] == '?') { - size_t j = i + 2; - while (j < in_len && input[j] != '>') j++; - i = (j < in_len) ? j + 1 : in_len; - continue; - } - int is_close = 0; - size_t name_start = i + 1; - if (input[i + 1] == '/') { - is_close = 1; - name_start = i + 2; - } - if (name_start >= in_len) { - html_buf_puts(&out, "<"); - i++; - continue; - } - unsigned char nc = (unsigned char)input[name_start]; - if (!((nc >= 'a' && nc <= 'z') || (nc >= 'A' && nc <= 'Z'))) { - /* `<` followed by non-letter — emit as escaped text. */ - html_buf_puts(&out, "<"); - i++; - continue; - } - size_t name_end = name_start; - while (name_end < in_len) { - unsigned char x = (unsigned char)input[name_end]; - if ((x >= 'a' && x <= 'z') || (x >= 'A' && x <= 'Z') || - (x >= '0' && x <= '9') || x == '-' || x == '_' || x == ':') { - name_end++; - } else { - break; - } - } - const char* tag = input + name_start; - size_t tag_len = name_end - name_start; - /* Find the `>` that closes this tag, respecting quoted attrs. */ - size_t cur = name_end; - int self_close = 0; - while (cur < in_len) { - unsigned char x = (unsigned char)input[cur]; - if (x == '"' || x == '\'') { - unsigned char q = x; - cur++; - while (cur < in_len && (unsigned char)input[cur] != q) cur++; - if (cur < in_len) cur++; /* skip closing quote */ - continue; - } - if (x == '/' && cur + 1 < in_len && input[cur + 1] == '>') { - self_close = 1; - break; - } - if (x == '>') break; - cur++; - } - if (cur >= in_len) { - /* Malformed: unclosed tag at EOF. Drop the rest of the input. */ - i = in_len; - continue; - } - size_t tag_end = self_close ? cur + 2 : cur + 1; /* one past `>` */ - /* Dangerous container — drop the whole subtree. */ - if (!is_close && html_is_dangerous_container(tag, tag_len)) { - if (self_close || html_is_void(tag, tag_len)) { - i = tag_end; - continue; - } - size_t scan = tag_end; - int found_close = 0; - while (scan < in_len) { - if (input[scan] != '<') { scan++; continue; } - if (scan + 1 < in_len && input[scan + 1] == '/') { - size_t cn_start = scan + 2; - size_t cn_end = cn_start; - while (cn_end < in_len) { - unsigned char x = (unsigned char)input[cn_end]; - if ((x >= 'a' && x <= 'z') || (x >= 'A' && x <= 'Z') || - (x >= '0' && x <= '9') || x == '-' || x == '_' || x == ':') { - cn_end++; - } else break; - } - if (cn_end - cn_start == tag_len && - html_iemem(input + cn_start, tag, tag_len)) { - size_t end_close = cn_end; - while (end_close < in_len && input[end_close] != '>') end_close++; - i = (end_close < in_len) ? end_close + 1 : in_len; - found_close = 1; - break; - } - } - scan++; - } - if (!found_close) { - /* No matching close — drop everything from here on. */ - i = in_len; - } - continue; - } - /* Look up the tag in the allowlist. */ - size_t arr_len = 0; - const char* arr = html_allowlist_find(allow, tag, tag_len, &arr_len); - if (!arr) { - /* Tag not allowed. Drop the open/close marker; inner text is - * processed by the outer loop and re-emitted as escaped text. */ - i = tag_end; - continue; - } - if (is_close) { - if (!html_is_void(tag, tag_len)) { - html_buf_putc(&out, '<'); - html_buf_putc(&out, '/'); - for (size_t k = 0; k < tag_len; k++) { - html_buf_putc(&out, (char)html_tolower((unsigned char)tag[k])); - } - html_buf_putc(&out, '>'); - } - i = tag_end; - continue; - } - /* Allowed open tag. Emit ``. */ - html_buf_putc(&out, '<'); - for (size_t k = 0; k < tag_len; k++) { - html_buf_putc(&out, (char)html_tolower((unsigned char)tag[k])); - } - size_t a = name_end; - while (a < cur) { - unsigned char x = (unsigned char)input[a]; - if (x == ' ' || x == '\t' || x == '\n' || x == '\r' || x == '/') { a++; continue; } - size_t an_start = a; - while (a < cur) { - unsigned char y = (unsigned char)input[a]; - if (y == '=' || y == ' ' || y == '\t' || y == '\n' || y == '\r' || y == '/' || y == '>') break; - a++; - } - size_t an_len = a - an_start; - if (an_len == 0) { a++; continue; } - size_t av_start = 0; - size_t av_len = 0; - int has_value = 0; - size_t b = a; - while (b < cur && (input[b] == ' ' || input[b] == '\t' || input[b] == '\n' || input[b] == '\r')) b++; - if (b < cur && input[b] == '=') { - has_value = 1; - b++; - while (b < cur && (input[b] == ' ' || input[b] == '\t' || input[b] == '\n' || input[b] == '\r')) b++; - if (b < cur && (input[b] == '"' || input[b] == '\'')) { - unsigned char q = (unsigned char)input[b]; - b++; - av_start = b; - while (b < cur && (unsigned char)input[b] != q) b++; - av_len = b - av_start; - if (b < cur) b++; - } else { - av_start = b; - while (b < cur) { - unsigned char y = (unsigned char)input[b]; - if (y == ' ' || y == '\t' || y == '\n' || y == '\r' || y == '>') break; - b++; - } - av_len = b - av_start; - } - a = b; - } - if (!html_attr_in_array(arr, arr_len, input + an_start, an_len)) continue; - int is_href = (an_len == 4 && html_iemem(input + an_start, "href", 4)); - int is_src = (an_len == 3 && html_iemem(input + an_start, "src", 3)); - if ((is_href || is_src) && has_value) { - if (!html_url_is_safe(input + av_start, av_len)) continue; - } - html_buf_putc(&out, ' '); - for (size_t k = 0; k < an_len; k++) { - html_buf_putc(&out, (char)html_tolower((unsigned char)input[an_start + k])); - } - if (has_value) { - html_buf_puts(&out, "=\""); - for (size_t k = 0; k < av_len; k++) { - unsigned char y = (unsigned char)input[av_start + k]; - /* Re-escape so the emitted attribute is well-formed - * double-quoted HTML. `&` passes through to preserve - * pre-encoded entities. */ - if (y == '"') html_buf_puts(&out, """); - else if (y == '<') html_buf_puts(&out, "<"); - else if (y == '>') html_buf_puts(&out, ">"); - else html_buf_putc(&out, (char)y); - } - html_buf_putc(&out, '"'); - } - } - html_buf_putc(&out, '>'); - i = tag_end; - } - /* Copy into arena-tracked buffer so the standard runtime memory model - * applies to the returned string. */ - char* result = el_strbuf(out.len); - memcpy(result, out.data, out.len); - result[out.len] = '\0'; - html_buf_free(&out); - return el_wrap_str(result); -} - -/* ── JSON ────────────────────────────────────────────────────────────────── */ - -/* True iff the segment is non-empty and every byte is an ASCII digit. We treat - * such segments as numeric array indices when walking a dot-path; mixed names - * like "0a" remain object-key lookups, so a key named "0" still wins over an - * index when the surrounding container is an object. */ -static int json_path_seg_is_index(const char* seg, size_t n) { - if (n == 0) return 0; - for (size_t i = 0; i < n; i++) { - if (seg[i] < '0' || seg[i] > '9') return 0; - } - return 1; -} - -/* Skip JSON whitespace. */ -static const char* json_skip_ws(const char* p) { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - return p; -} - -/* Descend one segment into the JSON cursor `p`. - * - If `p` points at an array `[...]` and the segment is all digits, - * advance to that element (zero-based). - * - Otherwise treat the segment as an object key and use json_find_key - * scoped to a one-level slice of the current container. - * Returns NULL if the descent fails (segment not found, container mismatch). - * - * `seg` is a pointer into the original path string and `seg_len` is its - * byte length — this avoids an extra alloc per segment. */ -static const char* json_path_descend(const char* p, const char* seg, size_t seg_len) { - if (!p || !seg) return NULL; - p = json_skip_ws(p); - if (*p == '[' && json_path_seg_is_index(seg, seg_len)) { - long idx = 0; - for (size_t i = 0; i < seg_len; i++) idx = idx * 10 + (seg[i] - '0'); - p++; /* step past '[' */ - p = json_skip_ws(p); - long cur = 0; - while (*p && *p != ']') { - if (cur == idx) return p; - const char* end = json_skip_value(p); - if (!end || end == p) return NULL; - p = json_skip_ws(end); - if (*p == ',') { p++; p = json_skip_ws(p); cur++; continue; } - /* No comma after this element — only acceptable at the closing ']', - * which means we ran out of elements. */ - break; - } - return NULL; - } - /* Object lookup. json_find_key walks at depth 1 of whatever container it - * receives, so we slice from `p` onwards. Caller already positioned us at - * the opening '{' (or at whitespace before it). */ - if (*p != '{') return NULL; - /* Build a NUL-terminated copy of the key segment for the lookup. We only - * pay this cost when the segment isn't a numeric index. */ - char stack_key[256]; - char* k = stack_key; - if (seg_len + 1 > sizeof(stack_key)) { - k = malloc(seg_len + 1); - if (!k) return NULL; - } - memcpy(k, seg, seg_len); - k[seg_len] = '\0'; - const char* found = json_find_key(p, k); - if (k != stack_key) free(k); - return found; -} - -/* Read the JSON value at `p` into a freshly-allocated, arena-owned el_val_t. - * - String -> unescaped, wrapped el_val_t string - * - Anything else -> raw JSON slice as a string (matches the historical - * json_get behaviour: numbers/bools/null come back stringified). */ -static el_val_t json_read_value(const char* p) { - p = json_skip_ws(p); - if (*p == '"') { - p++; - size_t cap = strlen(p) + 1; - char* out = el_strbuf(cap); - char* w = out; - while (*p && *p != '"') { - if (*p == '\\' && *(p+1)) { - p++; - switch (*p) { - case '"': *w++ = '"'; break; - case '\\': *w++ = '\\'; break; - case '/': *w++ = '/'; break; - case 'n': *w++ = '\n'; break; - case 'r': *w++ = '\r'; break; - case 't': *w++ = '\t'; break; - default: *w++ = *p; break; - } - } else { - *w++ = *p; - } - p++; - } - *w = '\0'; - return el_wrap_str(out); - } - /* Object/array/number/bool/null — return the raw slice up to the value's - * end. json_skip_value tracks brace/bracket/string state so nested objects - * round-trip cleanly. */ - const char* end = json_skip_value(p); - if (!end) end = p; - size_t n = (size_t)(end - p); - /* Strip trailing whitespace from scalar values so callers don't see - * `123 ` when they parsed a pretty-printed number. */ - while (n > 0 && (p[n-1] == ' ' || p[n-1] == '\t' || p[n-1] == '\n' || p[n-1] == '\r')) { - n--; - } - char* out = el_strbuf(n); - memcpy(out, p, n); - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t json_get(el_val_t jsonv, el_val_t keyv) { - const char* json = EL_CSTR(jsonv); - const char* key = EL_CSTR(keyv); - if (!json || !key) return el_wrap_str(el_strdup("")); - - /* Fast path: key contains no '.' — keep the historical single-segment - * substring search so existing callers retain their O(strlen) cost - * profile. The dot-path walker is only paid for when needed. */ - if (!strchr(key, '.')) { - size_t klen = strlen(key); - char stack_pat[512]; - char* pattern; - if (klen + 5 <= sizeof(stack_pat)) { - pattern = stack_pat; - } else { - pattern = malloc(klen + 5); - if (!pattern) return el_wrap_str(el_strdup("")); - } - snprintf(pattern, klen + 5, "\"%s\":", key); - const char* p = strstr(json, pattern); - if (pattern != stack_pat) free(pattern); - if (!p) return el_wrap_str(el_strdup("")); - p += strlen(key) + 3; /* skip "key": */ - return json_read_value(p); - } - - /* Dot-path traversal. Walk segments left to right; at each step, descend - * into the current container by either array index (all-digit segment on - * an array cursor) or object key. */ - const char* cursor = json_skip_ws(json); - const char* seg_start = key; - const char* k = key; - while (1) { - if (*k == '.' || *k == '\0') { - size_t seg_len = (size_t)(k - seg_start); - cursor = json_path_descend(cursor, seg_start, seg_len); - if (!cursor) return el_wrap_str(el_strdup("")); - if (*k == '\0') break; - k++; - seg_start = k; - continue; - } - k++; - } - return json_read_value(cursor); -} - -/* ── Float bit-cast helpers ──────────────────────────────────────────────── */ -/* `el_to_float` and `el_from_float` are exposed in el_runtime.h as static - * inlines so generated programs (which #include the header) can call them - * for Float literals. No definitions are needed here. */ - -/* ── JSON parser (recursive descent) ─────────────────────────────────────── */ -/* - * Parsed JSON representation: - * - object -> ElMap (keys & values are el_val_t) - * - array -> ElList - * - string -> EL_STR-wrapped char* (allocated) - * - number -> int (el_val_t) if integer, otherwise el_from_float(double) - * - true -> 1 - * - false -> 0 - * - null -> EL_NULL (0) - * - * Note: there is no runtime type tag — parsed numbers cannot be - * distinguished from booleans by the runtime alone. The codegen tracks - * types separately. This matches the rest of el_val_t's type-erased model. - */ - -/* JsonParser struct is forward-declared near the HTTP/Engram section. */ - -static void jp_skip_ws(JsonParser* jp) { - while (jp->p < jp->end) { - char c = *jp->p; - if (c == ' ' || c == '\t' || c == '\n' || c == '\r') jp->p++; - else break; - } -} - -static el_val_t jp_parse_value(JsonParser* jp); - -/* Parse a JSON string literal (the opening " has NOT yet been consumed). */ -static char* jp_parse_string_raw(JsonParser* jp) { - if (jp->p >= jp->end || *jp->p != '"') { jp->err = 1; return el_strdup(""); } - jp->p++; - size_t cap = 32, len = 0; - char* out = malloc(cap); - if (!out) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - while (jp->p < jp->end && *jp->p != '"') { - char c = *jp->p++; - if (c == '\\' && jp->p < jp->end) { - char esc = *jp->p++; - switch (esc) { - case '"': c = '"'; break; - case '\\': c = '\\'; break; - case '/': c = '/'; break; - case 'b': c = '\b'; break; - case 'f': c = '\f'; break; - case 'n': c = '\n'; break; - case 'r': c = '\r'; break; - case 't': c = '\t'; break; - case 'u': { - /* Skip 4 hex digits; emit '?' as a placeholder */ - for (int i = 0; i < 4 && jp->p < jp->end; i++) jp->p++; - c = '?'; - break; - } - default: c = esc; break; - } - } - if (len + 1 >= cap) { - cap *= 2; - out = realloc(out, cap); - if (!out) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - } - out[len++] = c; - } - if (jp->p < jp->end && *jp->p == '"') jp->p++; - else jp->err = 1; - out[len] = '\0'; - return out; -} - -static el_val_t jp_parse_number(JsonParser* jp) { - const char* start = jp->p; - int is_float = 0; - if (jp->p < jp->end && (*jp->p == '-' || *jp->p == '+')) jp->p++; - while (jp->p < jp->end && isdigit((unsigned char)*jp->p)) jp->p++; - if (jp->p < jp->end && *jp->p == '.') { - is_float = 1; jp->p++; - while (jp->p < jp->end && isdigit((unsigned char)*jp->p)) jp->p++; - } - if (jp->p < jp->end && (*jp->p == 'e' || *jp->p == 'E')) { - is_float = 1; jp->p++; - if (jp->p < jp->end && (*jp->p == '+' || *jp->p == '-')) jp->p++; - while (jp->p < jp->end && isdigit((unsigned char)*jp->p)) jp->p++; - } - size_t n = (size_t)(jp->p - start); - char buf[64]; - if (n >= sizeof(buf)) n = sizeof(buf) - 1; - memcpy(buf, start, n); - buf[n] = '\0'; - if (is_float) return el_from_float(strtod(buf, NULL)); - return (el_val_t)strtoll(buf, NULL, 10); -} - -static el_val_t jp_parse_array(JsonParser* jp) { - if (jp->p < jp->end && *jp->p == '[') jp->p++; - el_val_t lst = el_list_empty(); - jp_skip_ws(jp); - if (jp->p < jp->end && *jp->p == ']') { jp->p++; return lst; } - while (jp->p < jp->end) { - jp_skip_ws(jp); - el_val_t v = jp_parse_value(jp); - lst = el_list_append(lst, v); - jp_skip_ws(jp); - if (jp->p < jp->end && *jp->p == ',') { jp->p++; continue; } - if (jp->p < jp->end && *jp->p == ']') { jp->p++; break; } - jp->err = 1; - break; - } - return lst; -} - -static el_val_t jp_parse_object(JsonParser* jp) { - if (jp->p < jp->end && *jp->p == '{') jp->p++; - el_val_t m = el_map_new(0); - jp_skip_ws(jp); - if (jp->p < jp->end && *jp->p == '}') { jp->p++; return m; } - while (jp->p < jp->end) { - jp_skip_ws(jp); - char* key = jp_parse_string_raw(jp); - jp_skip_ws(jp); - if (jp->p < jp->end && *jp->p == ':') jp->p++; - else { jp->err = 1; free(key); break; } - jp_skip_ws(jp); - el_val_t v = jp_parse_value(jp); - m = el_map_set(m, EL_STR(key), v); - jp_skip_ws(jp); - if (jp->p < jp->end && *jp->p == ',') { jp->p++; continue; } - if (jp->p < jp->end && *jp->p == '}') { jp->p++; break; } - jp->err = 1; - break; - } - return m; -} - -static el_val_t jp_parse_value(JsonParser* jp) { - jp_skip_ws(jp); - if (jp->p >= jp->end) { jp->err = 1; return EL_NULL; } - char c = *jp->p; - if (c == '"') return el_wrap_str(jp_parse_string_raw(jp)); - if (c == '{') return jp_parse_object(jp); - if (c == '[') return jp_parse_array(jp); - if (c == '-' || isdigit((unsigned char)c)) return jp_parse_number(jp); - if (c == 't' && jp->p + 4 <= jp->end && strncmp(jp->p, "true", 4) == 0) { jp->p += 4; return 1; } - if (c == 'f' && jp->p + 5 <= jp->end && strncmp(jp->p, "false", 5) == 0) { jp->p += 5; return 0; } - if (c == 'n' && jp->p + 4 <= jp->end && strncmp(jp->p, "null", 4) == 0) { jp->p += 4; return EL_NULL; } - jp->err = 1; - return EL_NULL; -} - -el_val_t json_parse(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return EL_NULL; - JsonParser jp = { .p = s, .end = s + strlen(s), .err = 0 }; - el_val_t v = jp_parse_value(&jp); - if (jp.err) return EL_NULL; - return v; -} - -/* ── JSON stringify ──────────────────────────────────────────────────────── */ -/* - * Stringify policy: el_val_t is type-erased, so we cannot perfectly - * round-trip arbitrary values. We use these heuristics: - * - If value is an ElList pointer (in the heap range), serialize as array. - * - If value is an ElMap pointer, serialize as object. - * - If value looks like a printable string pointer, serialize as string. - * - Otherwise serialize as integer. - * This is best-effort. Programs that need exact control should build the - * string directly. A pointer test is the cheapest way to disambiguate - * from small integers without a separate type tag. - */ - -/* JsonBuf struct is forward-declared near the HTTP section so HTTP helpers - * can use it. Its definition appears there. */ - -static void jb_init(JsonBuf* b) { - b->cap = 64; b->len = 0; - b->buf = malloc(b->cap); - if (!b->buf) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - b->buf[0] = '\0'; -} - -static void jb_reserve(JsonBuf* b, size_t add) { - if (b->len + add + 1 > b->cap) { - while (b->len + add + 1 > b->cap) b->cap *= 2; - b->buf = realloc(b->buf, b->cap); - if (!b->buf) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - } -} - -static void jb_putc(JsonBuf* b, char c) { - jb_reserve(b, 1); - b->buf[b->len++] = c; - b->buf[b->len] = '\0'; -} - -static void jb_puts(JsonBuf* b, const char* s) { - size_t n = strlen(s); - jb_reserve(b, n); - memcpy(b->buf + b->len, s, n); - b->len += n; - b->buf[b->len] = '\0'; -} - -static void jb_emit_escaped(JsonBuf* b, const char* s) { - jb_putc(b, '"'); - for (; *s; s++) { - unsigned char c = (unsigned char)*s; - switch (c) { - case '"': jb_puts(b, "\\\""); break; - case '\\': jb_puts(b, "\\\\"); break; - case '\b': jb_puts(b, "\\b"); break; - case '\f': jb_puts(b, "\\f"); break; - case '\n': jb_puts(b, "\\n"); break; - case '\r': jb_puts(b, "\\r"); break; - case '\t': jb_puts(b, "\\t"); break; - default: - if (c < 0x20) { - char tmp[8]; - snprintf(tmp, sizeof(tmp), "\\u%04x", c); - jb_puts(b, tmp); - } else { - jb_putc(b, (char)c); - } - break; - } - } - jb_putc(b, '"'); -} - -/* Heuristic: is this el_val_t likely a pointer to an ElList? - * We can't fully verify, but pointers are large addresses, integers small. - * Treat values whose magnitude exceeds 2^32 as potential pointers and - * sniff by reading the header conservatively. - * - * Simpler heuristic: if the value reads as a printable string, treat as - * string; otherwise as integer. Lists/Maps are encoded as struct pointers, - * which have leading binary bytes — so they won't look like strings. */ - -static int looks_like_string(el_val_t v) { - if (v == 0) return 0; - /* Treat plausible heap addresses as candidates. - * Threshold: 4 GiB (0x100000000). On 64-bit systems heap addresses from - * malloc/mmap start well above 4 GiB (ASLR pushes them to ~0x7f...). - * El integer values (counters, unix timestamps up to ~2106) all fit below - * 0x100000000 (4294967296). The old threshold of 1,000,000 caused unix - * timestamps (~1.7e9) to be misidentified as string pointers — a segfault - * risk in json_stringify and jb_emit_value. */ - uintptr_t p = (uintptr_t)v; - if (p < 0x100000000ULL) return 0; /* integers, timestamps, counters */ - if (p < 0x1000) return 0; - /* Sniff first bytes for printable */ - const unsigned char* s = (const unsigned char*)p; - for (int i = 0; i < 16; i++) { - unsigned char c = s[i]; - if (c == '\0') return 1; /* terminated string (empty string is still a valid string) */ - /* Reject C0 control chars (non-whitespace), allow UTF-8 high bytes. - * 0x09-0x0d = tab/newline/cr/vt/ff (whitespace, OK) - * 0x20-0x7e = printable ASCII (OK) - * 0x7f = DEL (reject) - * 0x80-0xff = UTF-8 continuation/lead bytes (OK for multi-byte chars) */ - if (c < 0x09 || (c > 0x0d && c < 0x20) || c == 0x7f) return 0; - } - return 1; /* 16+ printable bytes — call it a string */ -} - -static void jb_emit_value(JsonBuf* b, el_val_t v); - -static void jb_emit_int(JsonBuf* b, int64_t n) { - char tmp[32]; - snprintf(tmp, sizeof(tmp), "%lld", (long long)n); - jb_puts(b, tmp); -} - -static void jb_emit_value(JsonBuf* b, el_val_t v) { - if (v == EL_NULL) { jb_puts(b, "null"); return; } - if (looks_like_string(v)) { - jb_emit_escaped(b, EL_CSTR(v)); - return; - } - jb_emit_int(b, (int64_t)v); -} - -el_val_t json_stringify(el_val_t v) { - JsonBuf b; jb_init(&b); - jb_emit_value(&b, v); - return el_wrap_str(b.buf); -} - -/* ── JSON substring accessors ────────────────────────────────────────────── */ -/* - * These walk the raw JSON string looking for "key": at the top level (depth 1) - * of an object. They handle escaped quotes, nested objects/arrays, and - * whitespace around the colon. - */ - -/* Find "key": at object-depth == 1 inside the JSON object string `s`. - * Returns pointer to the first byte of the value, or NULL. */ -static const char* json_find_key(const char* s, const char* key) { - if (!s || !key) return NULL; - size_t klen = strlen(key); - int depth = 0; - int in_str = 0; - int escape = 0; - const char* p = s; - while (*p) { - char c = *p; - if (in_str) { - if (escape) { escape = 0; } - else if (c == '\\') { escape = 1; } - else if (c == '"') { - /* End of string. If we're at depth 1, check if this was a key. */ - p++; - if (depth == 1) { - /* The string just ended at p-1. Check if it matches key - * and is followed by a colon. We need to backtrack to find - * the start of this string and compare. */ - } - in_str = 0; - continue; - } - p++; - continue; - } - if (c == '"') { - /* Start of a string literal */ - const char* str_start = p + 1; - const char* q = str_start; - int e = 0; - while (*q) { - if (e) { e = 0; q++; continue; } - if (*q == '\\') { e = 1; q++; continue; } - if (*q == '"') break; - q++; - } - size_t slen = (size_t)(q - str_start); - const char* after = (*q == '"') ? q + 1 : q; - /* If at depth 1 and matches key and followed by ':' -> got it */ - if (depth == 1 && slen == klen && strncmp(str_start, key, klen) == 0) { - const char* r = after; - while (*r == ' ' || *r == '\t' || *r == '\n' || *r == '\r') r++; - if (*r == ':') { - r++; - while (*r == ' ' || *r == '\t' || *r == '\n' || *r == '\r') r++; - return r; - } - } - p = after; - continue; - } - if (c == '{' || c == '[') depth++; - else if (c == '}' || c == ']') depth--; - p++; - } - return NULL; -} - -/* Skip a JSON value starting at p; return pointer past the value end. */ -static const char* json_skip_value(const char* p) { - if (!p || !*p) return p; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p == '"') { - p++; - int e = 0; - while (*p) { - if (e) { e = 0; p++; continue; } - if (*p == '\\') { e = 1; p++; continue; } - if (*p == '"') { p++; break; } - p++; - } - return p; - } - if (*p == '{' || *p == '[') { - char open = *p; - char close = (open == '{') ? '}' : ']'; - int depth = 0; - int in_str = 0; - int e = 0; - while (*p) { - char c = *p; - if (in_str) { - if (e) { e = 0; } - else if (c == '\\') { e = 1; } - else if (c == '"') in_str = 0; - p++; - continue; - } - if (c == '"') { in_str = 1; p++; continue; } - if (c == open) depth++; - else if (c == close) { depth--; p++; if (depth == 0) return p; continue; } - p++; - } - return p; - } - /* scalar: number, true/false/null */ - while (*p && *p != ',' && *p != '}' && *p != ']' && - *p != ' ' && *p != '\t' && *p != '\n' && *p != '\r') p++; - return p; -} - -el_val_t json_get_string(el_val_t json_str, el_val_t key) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - const char* p = json_find_key(json, k); - if (!p || *p != '"') return el_wrap_str(el_strdup("")); - p++; - JsonParser jp = { .p = p - 1, .end = json + (json ? strlen(json) : 0), .err = 0 }; - char* parsed = jp_parse_string_raw(&jp); - if (jp.err) { free(parsed); return el_wrap_str(el_strdup("")); } - return el_wrap_str(parsed); -} - -el_val_t json_get_int(el_val_t json_str, el_val_t key) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - const char* p = json_find_key(json, k); - if (!p) return 0; - if (*p == '"' || *p == '{' || *p == '[') return 0; - return (el_val_t)strtoll(p, NULL, 10); -} - -el_val_t json_get_float(el_val_t json_str, el_val_t key) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - const char* p = json_find_key(json, k); - if (!p) return 0; - if (*p == '"' || *p == '{' || *p == '[') return 0; - return el_from_float(strtod(p, NULL)); -} - -el_val_t json_get_bool(el_val_t json_str, el_val_t key) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - const char* p = json_find_key(json, k); - if (!p) return 0; - if (strncmp(p, "true", 4) == 0) return 1; - return 0; -} - -el_val_t json_get_raw(el_val_t json_str, el_val_t key) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - const char* p = json_find_key(json, k); - /* Clear fs_read binary-length hint — result is a fresh null-terminated - * string, not the raw file bytes, so Content-Length must use strlen. */ - _tl_fs_read_len = 0; - if (!p) return el_wrap_str(el_strdup("")); - const char* end = json_skip_value(p); - size_t n = (size_t)(end - p); - char* out = el_strbuf(n); - memcpy(out, p, n); - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t json_set(el_val_t json_str, el_val_t key, el_val_t value) { - const char* json = EL_CSTR(json_str); - const char* k = EL_CSTR(key); - if (!k) k = ""; - if (!json || !*json) { - /* Build a fresh object */ - JsonBuf b; jb_init(&b); - jb_putc(&b, '{'); - jb_emit_escaped(&b, k); - jb_putc(&b, ':'); - jb_emit_value(&b, value); - jb_putc(&b, '}'); - return el_wrap_str(b.buf); - } - const char* existing = json_find_key(json, k); - JsonBuf b; jb_init(&b); - if (existing) { - const char* end = json_skip_value(existing); - /* Copy [json .. existing) */ - size_t prefix = (size_t)(existing - json); - jb_reserve(&b, prefix); - memcpy(b.buf + b.len, json, prefix); - b.len += prefix; - b.buf[b.len] = '\0'; - jb_emit_value(&b, value); - jb_puts(&b, end); - return el_wrap_str(b.buf); - } - /* Insert before closing '}'. Find last '}' */ - size_t jl = strlen(json); - if (jl == 0) { free(b.buf); return el_wrap_str(el_strdup("{}")); } - /* Find last '}' from the end */ - ssize_t close_idx = -1; - for (ssize_t i = (ssize_t)jl - 1; i >= 0; i--) { - if (json[i] == '}') { close_idx = i; break; } - } - if (close_idx < 0) { - free(b.buf); - return el_wrap_str(el_strdup(json)); - } - /* Determine if object is empty: scan between last '{' and '}' for non-ws */ - int empty = 1; - for (ssize_t i = close_idx - 1; i >= 0; i--) { - char c = json[i]; - if (c == '{') break; - if (c != ' ' && c != '\t' && c != '\n' && c != '\r') { empty = 0; break; } - } - /* Copy json[0..close_idx) */ - jb_reserve(&b, (size_t)close_idx); - memcpy(b.buf + b.len, json, (size_t)close_idx); - b.len += (size_t)close_idx; - b.buf[b.len] = '\0'; - if (!empty) jb_putc(&b, ','); - jb_emit_escaped(&b, k); - jb_putc(&b, ':'); - jb_emit_value(&b, value); - /* Append from close_idx onward */ - jb_puts(&b, json + close_idx); - return el_wrap_str(b.buf); -} - -el_val_t json_array_len(el_val_t json_str) { - const char* s = EL_CSTR(json_str); - if (!s) return 0; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s != '[') return 0; - s++; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s == ']') return 0; - int64_t count = 0; - while (*s) { - const char* end = json_skip_value(s); - if (end == s) break; - count++; - s = end; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s == ',') { s++; continue; } - if (*s == ']' || *s == '\0') break; - } - return (el_val_t)count; -} - -/* json_array_get — return the i-th element of a JSON array as a JSON - * fragment string. Nested objects and arrays are returned verbatim - * (json_skip_value tracks brace/bracket depth so nested structures are - * preserved intact). Out-of-range index → "". */ -el_val_t json_array_get(el_val_t json_str, el_val_t index) { - const char* s = EL_CSTR(json_str); - int64_t idx = (int64_t)index; - if (!s || idx < 0) return el_wrap_str(el_strdup("")); - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s != '[') return el_wrap_str(el_strdup("")); - s++; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s == ']') return el_wrap_str(el_strdup("")); - int64_t i = 0; - while (*s) { - const char* start = s; - const char* end = json_skip_value(s); - if (end == s) break; - if (i == idx) { - size_t n = (size_t)(end - start); - char* out = el_strbuf(n); - memcpy(out, start, n); - out[n] = '\0'; - return el_wrap_str(out); - } - i++; - s = end; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s == ',') { s++; while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; continue; } - if (*s == ']' || *s == '\0') break; - } - return el_wrap_str(el_strdup("")); -} - -/* json_array_get_string — same as json_array_get, but assume the element - * is a JSON string and return the unquoted/unescaped value. Non-string - * elements yield "". */ -el_val_t json_array_get_string(el_val_t json_str, el_val_t index) { - el_val_t raw = json_array_get(json_str, index); - const char* s = EL_CSTR(raw); - if (!s || *s != '"') return el_wrap_str(el_strdup("")); - JsonParser jp = { - .p = s, - .end = s + strlen(s), - .err = 0, - }; - char* parsed = jp_parse_string_raw(&jp); - if (jp.err) { - free(parsed); - return el_wrap_str(el_strdup("")); - } - return el_wrap_str(parsed); -} - -/* ── Time ────────────────────────────────────────────────────────────────── */ - -el_val_t time_now(void) { - struct timeval tv; - gettimeofday(&tv, NULL); - int64_t ms = (int64_t)tv.tv_sec * 1000LL + (int64_t)tv.tv_usec / 1000LL; - return (el_val_t)ms; -} - -el_val_t time_now_utc(void) { - return time_now(); -} - -el_val_t time_format(el_val_t ts, el_val_t fmt) { - int64_t ms = (int64_t)ts; - time_t s = (time_t)(ms / 1000); - int msec = (int)(ms % 1000); - if (msec < 0) { msec += 1000; s -= 1; } - struct tm tm; - gmtime_r(&s, &tm); - const char* fmt_str = EL_CSTR(fmt); - if (!fmt_str || strcmp(fmt_str, "ISO") == 0) { - char buf[64]; - snprintf(buf, sizeof(buf), "%04d-%02d-%02dT%02d:%02d:%02d.%03dZ", - tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, - tm.tm_hour, tm.tm_min, tm.tm_sec, msec); - return el_wrap_str(el_strdup(buf)); - } - char buf[256]; - if (strftime(buf, sizeof(buf), fmt_str, &tm) == 0) buf[0] = '\0'; - return el_wrap_str(el_strdup(buf)); -} - -el_val_t time_to_parts(el_val_t ts) { - int64_t ms = (int64_t)ts; - time_t s = (time_t)(ms / 1000); - int msec = (int)(ms % 1000); - if (msec < 0) { msec += 1000; s -= 1; } - struct tm tm; - gmtime_r(&s, &tm); - el_val_t m = el_map_new(0); - m = el_map_set(m, EL_STR(el_strdup("year")), (el_val_t)(tm.tm_year + 1900)); - m = el_map_set(m, EL_STR(el_strdup("month")), (el_val_t)(tm.tm_mon + 1)); - m = el_map_set(m, EL_STR(el_strdup("day")), (el_val_t)tm.tm_mday); - m = el_map_set(m, EL_STR(el_strdup("hour")), (el_val_t)tm.tm_hour); - m = el_map_set(m, EL_STR(el_strdup("minute")), (el_val_t)tm.tm_min); - m = el_map_set(m, EL_STR(el_strdup("second")), (el_val_t)tm.tm_sec); - m = el_map_set(m, EL_STR(el_strdup("ms")), (el_val_t)msec); - return m; -} - -el_val_t time_from_parts(el_val_t secs, el_val_t ns, el_val_t tz) { - (void)tz; - int64_t s = (int64_t)secs; - int64_t n = (int64_t)ns; - int64_t ms = s * 1000LL + n / 1000000LL; - return (el_val_t)ms; -} - -el_val_t time_add(el_val_t ts, el_val_t n, el_val_t unit) { - const char* u = EL_CSTR(unit); - int64_t cur = (int64_t)ts; - int64_t d = (int64_t)n; - int64_t add_ms = d; - if (u) { - if (strcmp(u, "ms") == 0) add_ms = d; - else if (strcmp(u, "sec") == 0) add_ms = d * 1000LL; - else if (strcmp(u, "min") == 0) add_ms = d * 60000LL; - else if (strcmp(u, "hour") == 0) add_ms = d * 3600000LL; - else if (strcmp(u, "day") == 0) add_ms = d * 86400000LL; - } - return (el_val_t)(cur + add_ms); -} - -el_val_t time_diff(el_val_t ts1, el_val_t ts2, el_val_t unit) { - int64_t d = (int64_t)ts2 - (int64_t)ts1; - const char* u = EL_CSTR(unit); - if (!u || strcmp(u, "ms") == 0) return (el_val_t)d; - if (strcmp(u, "sec") == 0) return (el_val_t)(d / 1000LL); - if (strcmp(u, "min") == 0) return (el_val_t)(d / 60000LL); - if (strcmp(u, "hour") == 0) return (el_val_t)(d / 3600000LL); - if (strcmp(u, "day") == 0) return (el_val_t)(d / 86400000LL); - return (el_val_t)d; -} - -/* Block the calling thread for `secs` seconds. Negative values are clamped - * to 0. Used by El programs that poll external resources (e.g. RunPod - * /status, Engram readiness probes). */ -el_val_t sleep_secs(el_val_t secs) { - int64_t s = (int64_t)secs; - if (s < 0) s = 0; - struct timespec ts; - ts.tv_sec = (time_t)s; - ts.tv_nsec = 0; - nanosleep(&ts, NULL); - return 0; -} - -el_val_t sleep_ms(el_val_t ms) { - int64_t m = (int64_t)ms; - if (m < 0) m = 0; - struct timespec ts; - ts.tv_sec = (time_t)(m / 1000LL); - ts.tv_nsec = (long)((m % 1000LL) * 1000000LL); - nanosleep(&ts, NULL); - return 0; -} - -/* ── Instant + Duration: first-class temporal types ────────────────────────── - * El's substrate (Neuron) is a temporal cognition system. Memory salience - * decay, the six-tier pacemaker, TTL caches, and supersession are all - * temporal. Treating time as a raw Int (now() returning ms-since-epoch and - * arithmetic done with mixed unit literals) lets bugs through the type - * system: `(now - cached_at) < 60` cannot tell ms from sec, and `sleep(30)` - * is ambiguous. This block introduces two dedicated representations. - * - * Representation: - * Instant — int64 nanoseconds since the Unix epoch - * Duration — int64 nanoseconds (signed; negative durations are legal, - * e.g. when a deadline has passed) - * - * Both share the el_val_t (int64) slot the rest of the runtime uses, so no - * boxing / arena allocation is needed. Type discipline is enforced at the - * codegen layer: `let x: Duration = ...` registers `x` in __duration_names, - * and BinOp dispatches through typed wrappers (el_duration_add, etc.) that - * make intent explicit in the generated C. Mismatched ops (Instant+Instant, - * Duration+Int) are surfaced via #error directives at codegen time so the - * downstream cc step fails with a clear El-source-level message. - * - * Nanosecond precision matches POSIX clock_gettime / nanosleep granularity. - * 2^63 nanos covers ~292 years from epoch — comfortably past 2200, plenty - * for a memory-system runtime that never schedules outside a human lifespan. - */ - -/* now() — current Instant. Wraps clock_gettime(CLOCK_REALTIME) for nanosecond - * precision. Falls back to gettimeofday on systems where clock_gettime is - * unavailable (defensive — every supported platform has it). */ -el_val_t el_now_instant(void) { - struct timespec ts; - if (clock_gettime(CLOCK_REALTIME, &ts) == 0) { - int64_t ns = (int64_t)ts.tv_sec * 1000000000LL + (int64_t)ts.tv_nsec; - return (el_val_t)ns; - } - struct timeval tv; - gettimeofday(&tv, NULL); - int64_t ns = (int64_t)tv.tv_sec * 1000000000LL - + (int64_t)tv.tv_usec * 1000LL; - return (el_val_t)ns; -} - -el_val_t now(void) { - return el_now_instant(); -} - -/* unix_seconds(n) — Instant from a Unix-epoch second count. - * unix_millis(n) — Instant from a Unix-epoch millisecond count. */ -el_val_t unix_seconds(el_val_t n) { - int64_t s = (int64_t)n; - return (el_val_t)(s * 1000000000LL); -} - -el_val_t unix_millis(el_val_t n) { - int64_t m = (int64_t)n; - return (el_val_t)(m * 1000000LL); -} - -/* instant_from_iso8601 — parse a strict subset: - * YYYY-MM-DDTHH:MM:SS[.fff]Z - * Returns 0 (the Unix-epoch sentinel) on parse failure. Callers that need to - * distinguish epoch-zero from a parse error should use a wider sentinel - * representation; the current zero-on-failure choice matches existing El - * runtime conventions for parse builtins (str_to_int, parse_int). */ -el_val_t instant_from_iso8601(el_val_t s) { - const char* str = EL_CSTR(s); - if (!str) return (el_val_t)0; - int Y, M, D, h, m, sec, frac = 0; - int n = sscanf(str, "%d-%d-%dT%d:%d:%d.%3d", &Y, &M, &D, &h, &m, &sec, &frac); - if (n < 6) { - n = sscanf(str, "%d-%d-%dT%d:%d:%dZ", &Y, &M, &D, &h, &m, &sec); - if (n < 6) return (el_val_t)0; - } - struct tm tm; - memset(&tm, 0, sizeof(tm)); - tm.tm_year = Y - 1900; - tm.tm_mon = M - 1; - tm.tm_mday = D; - tm.tm_hour = h; - tm.tm_min = m; - tm.tm_sec = sec; - /* timegm — UTC. POSIX-Y but available on macOS and glibc. */ - time_t t = timegm(&tm); - if (t == (time_t)-1) return (el_val_t)0; - int64_t ns = (int64_t)t * 1000000000LL + (int64_t)frac * 1000000LL; - return (el_val_t)ns; -} - -/* Duration constructors. The El-side postfix literals (30.seconds, 1.hour) - * are lowered by the codegen directly into a literal int64 of nanoseconds — - * these constructors are for runtime values where the count is dynamic. */ -el_val_t el_duration_from_nanos(el_val_t ns) { - return (el_val_t)(int64_t)ns; -} - -el_val_t duration_seconds(el_val_t n) { - int64_t s = (int64_t)n; - return (el_val_t)(s * 1000000000LL); -} - -el_val_t duration_millis(el_val_t n) { - int64_t m = (int64_t)n; - return (el_val_t)(m * 1000000LL); -} - -el_val_t duration_nanos(el_val_t n) { - return (el_val_t)(int64_t)n; -} - -/* Arithmetic — typed wrappers. At the C level these are no-op casts, but - * the codegen routes Instant/Duration BinOps through them so the generated - * C says `el_instant_add_dur(start, dur)` rather than `start + dur`. The - * intent is explicit, the operand order is documented, and a future change - * to the underlying representation (saturating arithmetic, overflow guards) - * has a single chokepoint. */ -el_val_t el_instant_add_dur(el_val_t inst, el_val_t dur) { - return (el_val_t)((int64_t)inst + (int64_t)dur); -} - -el_val_t el_instant_sub_dur(el_val_t inst, el_val_t dur) { - return (el_val_t)((int64_t)inst - (int64_t)dur); -} - -el_val_t el_instant_diff(el_val_t a, el_val_t b) { - /* a - b — yields a Duration (negative if b is later than a). */ - return (el_val_t)((int64_t)a - (int64_t)b); -} - -el_val_t el_duration_add(el_val_t a, el_val_t b) { - return (el_val_t)((int64_t)a + (int64_t)b); -} - -el_val_t el_duration_sub(el_val_t a, el_val_t b) { - return (el_val_t)((int64_t)a - (int64_t)b); -} - -el_val_t el_duration_scale(el_val_t dur, el_val_t scalar) { - return (el_val_t)((int64_t)dur * (int64_t)scalar); -} - -el_val_t el_duration_div(el_val_t dur, el_val_t scalar) { - int64_t s = (int64_t)scalar; - if (s == 0) return (el_val_t)0; - return (el_val_t)((int64_t)dur / s); -} - -/* Comparisons. Return 1/0 in el_val_t convention. */ -el_val_t el_instant_lt(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a < (int64_t)b ? 1 : 0); } -el_val_t el_instant_le(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a <= (int64_t)b ? 1 : 0); } -el_val_t el_instant_gt(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a > (int64_t)b ? 1 : 0); } -el_val_t el_instant_ge(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a >= (int64_t)b ? 1 : 0); } -el_val_t el_instant_eq(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a == (int64_t)b ? 1 : 0); } -el_val_t el_instant_ne(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a != (int64_t)b ? 1 : 0); } -el_val_t el_duration_lt(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a < (int64_t)b ? 1 : 0); } -el_val_t el_duration_le(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a <= (int64_t)b ? 1 : 0); } -el_val_t el_duration_gt(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a > (int64_t)b ? 1 : 0); } -el_val_t el_duration_ge(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a >= (int64_t)b ? 1 : 0); } -el_val_t el_duration_eq(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a == (int64_t)b ? 1 : 0); } -el_val_t el_duration_ne(el_val_t a, el_val_t b) { return (el_val_t)((int64_t)a != (int64_t)b ? 1 : 0); } - -/* Conversions. */ -el_val_t instant_to_unix_seconds(el_val_t i) { - return (el_val_t)((int64_t)i / 1000000000LL); -} - -el_val_t instant_to_unix_millis(el_val_t i) { - return (el_val_t)((int64_t)i / 1000000LL); -} - -el_val_t instant_to_iso8601(el_val_t i) { - int64_t ns = (int64_t)i; - time_t s = (time_t)(ns / 1000000000LL); - int msec = (int)((ns / 1000000LL) % 1000LL); - if (msec < 0) { msec += 1000; s -= 1; } - struct tm tm; - gmtime_r(&s, &tm); - char buf[64]; - snprintf(buf, sizeof(buf), "%04d-%02d-%02dT%02d:%02d:%02d.%03dZ", - tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, - tm.tm_hour, tm.tm_min, tm.tm_sec, msec); - return el_wrap_str(el_strdup(buf)); -} - -el_val_t duration_to_seconds(el_val_t d) { - return (el_val_t)((int64_t)d / 1000000000LL); -} - -el_val_t duration_to_millis(el_val_t d) { - return (el_val_t)((int64_t)d / 1000000LL); -} - -el_val_t duration_to_nanos(el_val_t d) { - return (el_val_t)(int64_t)d; -} - -/* sleep(Duration) — Phase 1 replacement for ambiguous sleep(Int). The runtime - * still exposes sleep_secs/sleep_ms for legacy call sites; codegen lowers - * sleep(Duration) to el_sleep_duration(d). Negative durations clamp to 0 so a - * stale deadline doesn't block forever. */ -el_val_t el_sleep_duration(el_val_t dur) { - int64_t ns = (int64_t)dur; - if (ns < 0) ns = 0; - struct timespec ts; - ts.tv_sec = (time_t)(ns / 1000000000LL); - ts.tv_nsec = (long)(ns % 1000000000LL); - nanosleep(&ts, NULL); - return (el_val_t)0; -} - -/* unix_timestamp() — back-compat. Existing El callers expect an Int seconds - * value; this stays an Int returner so the type system isn't disturbed for - * legacy code. New code should call now() and convert when needed. */ -el_val_t unix_timestamp(void) { - return instant_to_unix_seconds(el_now_instant()); -} - -/* TTL cache helpers. Backed by the existing process-wide K/V (state_set/get) - * with a sibling __ttl_set_at_ entry recording the Instant of the last - * write. ttl_cache_get returns "" if the entry is missing or stale, so call - * sites can branch on `if v == "" { miss } else { hit }` — the same shape - * existing get-with-default code uses. No more (now - cached_at) < 60. */ -el_val_t ttl_cache_set(el_val_t key, el_val_t value) { - const char* k = EL_CSTR(key); - if (!k) return (el_val_t)0; - /* Store the value at the user's key. */ - state_set(key, value); - /* Stamp set_at — opaque schema, namespaced under __ttl: prefix so user - * keys can't collide with stamps. */ - size_t klen = strlen(k); - char* stamp_key = (char*)malloc(klen + 16); - if (!stamp_key) return (el_val_t)0; - snprintf(stamp_key, klen + 16, "__ttl_at:%s", k); - int64_t now_ns = (int64_t)el_now_instant(); - char buf[32]; - snprintf(buf, sizeof(buf), "%lld", (long long)now_ns); - state_set(EL_STR(stamp_key), EL_STR(buf)); - free(stamp_key); - return (el_val_t)1; -} - -el_val_t ttl_cache_get(el_val_t key, el_val_t max_age) { - const char* k = EL_CSTR(key); - if (!k) return el_wrap_str(el_strdup("")); - /* Look up stamp. */ - size_t klen = strlen(k); - char* stamp_key = (char*)malloc(klen + 16); - if (!stamp_key) return el_wrap_str(el_strdup("")); - snprintf(stamp_key, klen + 16, "__ttl_at:%s", k); - el_val_t stamp = state_get(EL_STR(stamp_key)); - free(stamp_key); - const char* sv = EL_CSTR(stamp); - if (!sv || !*sv) return el_wrap_str(el_strdup("")); - int64_t set_at = (int64_t)atoll(sv); - int64_t now_ns = (int64_t)el_now_instant(); - int64_t age = now_ns - set_at; - int64_t max_ns = (int64_t)max_age; - if (age < 0) return el_wrap_str(el_strdup("")); /* clock skew — treat as miss */ - if (age > max_ns) return el_wrap_str(el_strdup("")); /* expired */ - return state_get(key); -} - -el_val_t ttl_cache_age(el_val_t key) { - const char* k = EL_CSTR(key); - if (!k) return (el_val_t)INT64_MAX; - size_t klen = strlen(k); - char* stamp_key = (char*)malloc(klen + 16); - if (!stamp_key) return (el_val_t)INT64_MAX; - snprintf(stamp_key, klen + 16, "__ttl_at:%s", k); - el_val_t stamp = state_get(EL_STR(stamp_key)); - free(stamp_key); - const char* sv = EL_CSTR(stamp); - if (!sv || !*sv) return (el_val_t)INT64_MAX; - int64_t set_at = (int64_t)atoll(sv); - int64_t now_ns = (int64_t)el_now_instant(); - return (el_val_t)(now_ns - set_at); -} - -/* ── Calendar + CalendarTime + Rhythm + LocalDate/Time/DateTime ────────────── - * Phase 1.5. Calendar is pluggable: EarthCalendar (IANA zones + Gregorian + - * DST), MarsCalendar (sols, MTC), CycleCalendar(period), NoCycleCalendar, - * RelativeCalendar(epoch). Phase 1 zone wrapping folds INTO EarthCalendar; - * UTC and IANA zones are themselves Earth-parochial and cannot live at the - * lowest type layer. - * - * A Rhythm is a small AST that asks the Calendar for cycle phase, weekday, - * etc. Most rhythm logic is calendar-agnostic at runtime: rhythm_cycle_phase - * means "midpoint of cycle" whether the cycle is 24h on Earth or 30h on a - * station or 300y on a long-cycle world. */ - -/* Magic headers — used by the runtime to recognize boxed temporal values - * arriving through el_val_t. Distinct constants so accidental misuse fails - * loudly rather than silently. */ -#define EL_CAL_MAGIC 0xE1CA1EDDU -#define EL_CALTIME_MAGIC 0xE1CA1747U -#define EL_RHYTHM_MAGIC 0xE1287A11U -#define EL_LDATE_MAGIC 0xE1DA7E00U -#define EL_LDT_MAGIC 0xE1DA7E1DU -#define EL_ZONE_MAGIC 0xE12017E0U - -typedef enum { - EL_CALENDAR_EARTH = 1, - EL_CALENDAR_MARS = 2, - EL_CALENDAR_CYCLE = 3, - EL_CALENDAR_NO_CYCLE = 4, - EL_CALENDAR_RELATIVE = 5 -} el_calendar_kind_t; - -typedef struct { - uint32_t magic; - char* id; /* IANA name or "+HH:MM" / "-HH:MM" */ - int fixed; /* 1 for fixed offset, 0 for IANA */ - int64_t offset_ns; /* fixed offset in nanos (only when fixed) */ -} el_zone_t; - -typedef struct { - uint32_t magic; - el_calendar_kind_t kind; - el_zone_t* zone; /* EarthCalendar; MarsCalendar uses MTC */ - int64_t cycle_period_ns;/* CycleCalendar; computed for Earth (86400 s) and Mars (88775.244 s) */ - int64_t epoch_ns; /* RelativeCalendar; Unix-epoch zero otherwise */ -} el_calendar_t; - -typedef struct { - uint32_t magic; - int64_t instant_ns; - el_calendar_t* cal; -} el_caltime_t; - -/* Rhythm AST. */ -typedef enum { - EL_RHYTHM_CYCLE_START = 1, - EL_RHYTHM_CYCLE_PHASE = 2, - EL_RHYTHM_DURATION = 3, - EL_RHYTHM_SESSION_START = 4, - EL_RHYTHM_EVENT = 5, - EL_RHYTHM_AND = 6, - EL_RHYTHM_OR = 7, - EL_RHYTHM_WEEKDAY = 8, - EL_RHYTHM_WEEKLY_AT = 9 -} el_rhythm_kind_t; - -typedef struct el_rhythm_s { - uint32_t magic; - el_rhythm_kind_t kind; - double phase; /* CYCLE_PHASE */ - int64_t period_ns; /* DURATION */ - int weekday; /* 1..7 Mon..Sun */ - int hour; - int minute; - char* event_name; /* EVENT */ - struct el_rhythm_s* a; /* AND/OR */ - struct el_rhythm_s* b; -} el_rhythm_t; - -typedef struct { - uint32_t magic; - int year; - int month; - int day; -} el_localdate_t; - -typedef struct { - uint32_t magic; - el_localdate_t* date; - int64_t time_ns; /* nanos since midnight */ -} el_localdt_t; - -/* Magic-tag check helpers — peek the first 4 bytes of an el_val_t pointer - * and compare against the expected magic. Strings are NUL-terminated and - * never start with our magic byte sequence, so this is safe. */ -static int el_is_magic(el_val_t v, uint32_t want) { - if (v == 0) return 0; - /* Defensive: only follow pointers in plausible address space. - * On 64-bit unix processes pointers are above 0x10000. */ - if ((uint64_t)v < 0x10000ULL) return 0; - uint32_t got = *(volatile uint32_t*)(uintptr_t)v; - return got == want; -} - -/* Sol length on Mars in nanoseconds: 88775.244 seconds. */ -#define EL_MARS_SOL_NS ((int64_t)88775244000000LL) -/* Earth solar day in nanoseconds: 86400 seconds. */ -#define EL_EARTH_DAY_NS ((int64_t)86400000000000LL) - -/* ── Zone construction ────────────────────────────────────────────────────── - * Zones intern by id string so equality comparisons are pointer-compares. */ - -#define EL_ZONE_TABLE_CAP 64 -static el_zone_t* _el_zone_table[EL_ZONE_TABLE_CAP]; -static int _el_zone_count = 0; - -static el_zone_t* _el_zone_intern(const char* id, int fixed, int64_t offset_ns) { - for (int i = 0; i < _el_zone_count; i++) { - el_zone_t* z = _el_zone_table[i]; - if (z->fixed == fixed && z->offset_ns == offset_ns && - strcmp(z->id ? z->id : "", id ? id : "") == 0) { - return z; - } - } - if (_el_zone_count >= EL_ZONE_TABLE_CAP) { - /* Out of slots: build a non-interned zone. Equality will fail across - * such zones but the program still runs. */ - el_zone_t* z = (el_zone_t*)malloc(sizeof(el_zone_t)); - z->magic = EL_ZONE_MAGIC; - z->id = el_strdup_persist(id ? id : ""); - z->fixed = fixed; - z->offset_ns = offset_ns; - return z; - } - el_zone_t* z = (el_zone_t*)malloc(sizeof(el_zone_t)); - z->magic = EL_ZONE_MAGIC; - z->id = el_strdup_persist(id ? id : ""); - z->fixed = fixed; - z->offset_ns = offset_ns; - _el_zone_table[_el_zone_count++] = z; - return z; -} - -el_val_t zone(el_val_t id) { - const char* s = EL_CSTR(id); - if (!s || !*s) return (el_val_t)(uintptr_t)_el_zone_intern("UTC", 0, 0); - /* Fixed-offset shortcut: "+HH:MM" or "-HH:MM". */ - if ((s[0] == '+' || s[0] == '-') && strlen(s) >= 6 && s[3] == ':') { - int sign = (s[0] == '-') ? -1 : 1; - int hh = (s[1] - '0') * 10 + (s[2] - '0'); - int mm = (s[4] - '0') * 10 + (s[5] - '0'); - int64_t off = (int64_t)sign * ((int64_t)hh * 3600LL + (int64_t)mm * 60LL) * 1000000000LL; - return (el_val_t)(uintptr_t)_el_zone_intern(s, 1, off); - } - return (el_val_t)(uintptr_t)_el_zone_intern(s, 0, 0); -} - -el_val_t zone_utc(void) { - return (el_val_t)(uintptr_t)_el_zone_intern("UTC", 1, 0); -} - -el_val_t zone_local(void) { - /* Resolve the local zone via TZ env or system default. tzset() picks - * up TZ if set; otherwise the C library reads /etc/localtime. We store - * the zone id as "LOCAL" so subsequent equality holds; resolution is - * lazy at use time. */ - return (el_val_t)(uintptr_t)_el_zone_intern("LOCAL", 0, 0); -} - -el_val_t zone_offset(el_val_t hours, el_val_t minutes) { - int hh = (int)(int64_t)hours; - int mm = (int)(int64_t)minutes; - int sign = (hh < 0 || mm < 0) ? -1 : 1; - if (hh < 0) hh = -hh; - if (mm < 0) mm = -mm; - int64_t off = (int64_t)sign * ((int64_t)hh * 3600LL + (int64_t)mm * 60LL) * 1000000000LL; - char buf[16]; - snprintf(buf, sizeof(buf), "%c%02d:%02d", sign < 0 ? '-' : '+', hh, mm); - return (el_val_t)(uintptr_t)_el_zone_intern(buf, 1, off); -} - -/* ── Calendar interning ──────────────────────────────────────────────────── */ - -#define EL_CAL_TABLE_CAP 64 -static el_calendar_t* _el_cal_table[EL_CAL_TABLE_CAP]; -static int _el_cal_count = 0; - -static el_calendar_t* _el_cal_intern(el_calendar_kind_t kind, el_zone_t* z, - int64_t period_ns, int64_t epoch_ns) { - for (int i = 0; i < _el_cal_count; i++) { - el_calendar_t* c = _el_cal_table[i]; - if (c->kind == kind && c->zone == z && - c->cycle_period_ns == period_ns && c->epoch_ns == epoch_ns) { - return c; - } - } - el_calendar_t* c = (el_calendar_t*)malloc(sizeof(el_calendar_t)); - c->magic = EL_CAL_MAGIC; - c->kind = kind; - c->zone = z; - c->cycle_period_ns = period_ns; - c->epoch_ns = epoch_ns; - if (_el_cal_count < EL_CAL_TABLE_CAP) _el_cal_table[_el_cal_count++] = c; - return c; -} - -el_val_t earth_calendar(el_val_t z_val) { - el_zone_t* z = NULL; - if (z_val != 0 && el_is_magic(z_val, EL_ZONE_MAGIC)) { - z = (el_zone_t*)(uintptr_t)z_val; - } else { - z = (el_zone_t*)(uintptr_t)zone_local(); - } - return (el_val_t)(uintptr_t)_el_cal_intern(EL_CALENDAR_EARTH, z, EL_EARTH_DAY_NS, 0); -} - -el_val_t earth_calendar_default(void) { - return earth_calendar(zone_local()); -} - -el_val_t mars_calendar(void) { - el_zone_t* z = (el_zone_t*)(uintptr_t)_el_zone_intern("MTC", 1, 0); - return (el_val_t)(uintptr_t)_el_cal_intern(EL_CALENDAR_MARS, z, EL_MARS_SOL_NS, 0); -} - -el_val_t cycle_calendar(el_val_t period_dur) { - int64_t period = (int64_t)period_dur; - if (period <= 0) period = 1; - return (el_val_t)(uintptr_t)_el_cal_intern(EL_CALENDAR_CYCLE, NULL, period, 0); -} - -el_val_t no_cycle_calendar(void) { - return (el_val_t)(uintptr_t)_el_cal_intern(EL_CALENDAR_NO_CYCLE, NULL, 0, 0); -} - -el_val_t relative_calendar(el_val_t epoch_inst) { - int64_t ep = (int64_t)epoch_inst; - return (el_val_t)(uintptr_t)_el_cal_intern(EL_CALENDAR_RELATIVE, NULL, 0, ep); -} - -/* ── CalendarTime ───────────────────────────────────────────────────────── */ - -static el_caltime_t* _el_caltime_alloc(int64_t inst, el_calendar_t* c) { - el_caltime_t* ct = (el_caltime_t*)malloc(sizeof(el_caltime_t)); - ct->magic = EL_CALTIME_MAGIC; - ct->instant_ns = inst; - ct->cal = c; - return ct; -} - -static el_calendar_t* _el_resolve_cal(el_val_t cal_val) { - if (cal_val == 0 || !el_is_magic(cal_val, EL_CAL_MAGIC)) { - return (el_calendar_t*)(uintptr_t)earth_calendar_default(); - } - return (el_calendar_t*)(uintptr_t)cal_val; -} - -el_val_t now_in(el_val_t cal_val) { - el_calendar_t* c = _el_resolve_cal(cal_val); - int64_t ns = (int64_t)el_now_instant(); - return (el_val_t)(uintptr_t)_el_caltime_alloc(ns, c); -} - -el_val_t in_calendar(el_val_t inst, el_val_t cal_val) { - el_calendar_t* c = _el_resolve_cal(cal_val); - return (el_val_t)(uintptr_t)_el_caltime_alloc((int64_t)inst, c); -} - -el_val_t cal_to_instant(el_val_t ct_val) { - if (!el_is_magic(ct_val, EL_CALTIME_MAGIC)) return (el_val_t)0; - el_caltime_t* ct = (el_caltime_t*)(uintptr_t)ct_val; - return (el_val_t)ct->instant_ns; -} - -el_val_t cal_in(el_val_t ct_val, el_val_t cal_val) { - if (!el_is_magic(ct_val, EL_CALTIME_MAGIC)) return (el_val_t)0; - el_caltime_t* ct = (el_caltime_t*)(uintptr_t)ct_val; - el_calendar_t* c = _el_resolve_cal(cal_val); - return (el_val_t)(uintptr_t)_el_caltime_alloc(ct->instant_ns, c); -} - -el_val_t cal_cycle_phase(el_val_t ct_val) { - if (!el_is_magic(ct_val, EL_CALTIME_MAGIC)) return el_from_float(0.0); - el_caltime_t* ct = (el_caltime_t*)(uintptr_t)ct_val; - el_calendar_t* c = ct->cal; - if (c->kind == EL_CALENDAR_NO_CYCLE) { - return el_from_float(0.0/0.0); /* NaN sentinel */ - } - int64_t period = c->cycle_period_ns; - if (period <= 0) return el_from_float(0.0); - int64_t base = ct->instant_ns - c->epoch_ns; - int64_t phase_ns = base % period; - if (phase_ns < 0) phase_ns += period; - double phase = (double)phase_ns / (double)period; - return el_from_float(phase); -} - -/* ── Earth zone resolution: TZ-based offset lookup ────────────────────────── - * For an EarthCalendar(zone), we want to convert an instant_ns into local - * y/m/d/h/m/s, including DST. Approach: setenv("TZ", id), tzset(), use - * localtime_r, then restore. This is not thread-safe by design — El's - * runtime is single-threaded for the request handler path. Cache the - * computed (instant -> tm) to avoid the syscall churn on repeat formats. */ - -static void _el_apply_zone(el_zone_t* z) { - if (!z) { unsetenv("TZ"); tzset(); return; } - if (z->fixed && strcmp(z->id, "UTC") == 0) { - setenv("TZ", "UTC0", 1); - tzset(); - return; - } - if (z->fixed) { - /* Fixed offset: POSIX TZ uses inverted sign (sign convention of - * "hours WEST of UTC" rather than east). Build the spec accordingly. */ - char buf[32]; - int neg_secs = (int)(-z->offset_ns / 1000000000LL); - int sign = neg_secs < 0 ? -1 : 1; - int abs_secs = neg_secs < 0 ? -neg_secs : neg_secs; - int hh = abs_secs / 3600; - int mm = (abs_secs % 3600) / 60; - snprintf(buf, sizeof(buf), "FIX%c%d:%02d", sign < 0 ? '-' : '+', hh, mm); - setenv("TZ", buf, 1); - tzset(); - return; - } - if (strcmp(z->id, "LOCAL") == 0) { - unsetenv("TZ"); - tzset(); - return; - } - setenv("TZ", z->id, 1); - tzset(); -} - -static int _el_decompose_earth(el_caltime_t* ct, struct tm* tm_out, int* abbr_len, char* abbr_buf, size_t abbr_cap) { - el_calendar_t* c = ct->cal; - el_zone_t* z = c->zone; - _el_apply_zone(z); - time_t s = (time_t)(ct->instant_ns / 1000000000LL); - struct tm tm; - localtime_r(&s, &tm); - *tm_out = tm; - if (abbr_buf && abbr_cap > 0) { - const char* z_str = tm.tm_zone ? tm.tm_zone : ""; - size_t n = strlen(z_str); - if (n >= abbr_cap) n = abbr_cap - 1; - memcpy(abbr_buf, z_str, n); - abbr_buf[n] = '\0'; - if (abbr_len) *abbr_len = (int)n; - } - return 0; -} - -/* Format an Earth CalendarTime under a Java-DateTimeFormatter-ish pattern. - * We support a useful core: yyyy MM dd HH mm ss z EEE MMM d h a — enough for - * the acceptance tests. Single quotes denote literal text. */ -static const char* _el_weekday_short[] = {"Sun","Mon","Tue","Wed","Thu","Fri","Sat"}; -static const char* _el_month_short[] = {"Jan","Feb","Mar","Apr","May","Jun", - "Jul","Aug","Sep","Oct","Nov","Dec"}; - -static char* _el_format_earth(el_caltime_t* ct, const char* pattern) { - struct tm tm; - char abbr[16] = {0}; - int abbr_len = 0; - _el_decompose_earth(ct, &tm, &abbr_len, abbr, sizeof(abbr)); - size_t cap = strlen(pattern) * 4 + 64; - char* out = (char*)malloc(cap); - size_t pos = 0; - size_t i = 0; - size_t plen = strlen(pattern); - while (i < plen) { - char ch = pattern[i]; - /* Quoted literal */ - if (ch == '\'') { - i++; - while (i < plen && pattern[i] != '\'') { - if (pos + 1 >= cap) { cap *= 2; out = realloc(out, cap); } - out[pos++] = pattern[i++]; - } - if (i < plen) i++; - continue; - } - /* Count run of same letter */ - size_t run = 1; - while (i + run < plen && pattern[i + run] == ch) run++; - char tmp[64]; - tmp[0] = '\0'; - if (ch == 'y') { - if (run >= 4) snprintf(tmp, sizeof(tmp), "%04d", tm.tm_year + 1900); - else snprintf(tmp, sizeof(tmp), "%02d", (tm.tm_year + 1900) % 100); - } else if (ch == 'M') { - if (run >= 3) snprintf(tmp, sizeof(tmp), "%s", _el_month_short[tm.tm_mon]); - else if (run == 2) snprintf(tmp, sizeof(tmp), "%02d", tm.tm_mon + 1); - else snprintf(tmp, sizeof(tmp), "%d", tm.tm_mon + 1); - } else if (ch == 'd') { - if (run >= 2) snprintf(tmp, sizeof(tmp), "%02d", tm.tm_mday); - else snprintf(tmp, sizeof(tmp), "%d", tm.tm_mday); - } else if (ch == 'H') { - if (run >= 2) snprintf(tmp, sizeof(tmp), "%02d", tm.tm_hour); - else snprintf(tmp, sizeof(tmp), "%d", tm.tm_hour); - } else if (ch == 'h') { - int h12 = tm.tm_hour % 12; if (h12 == 0) h12 = 12; - if (run >= 2) snprintf(tmp, sizeof(tmp), "%02d", h12); - else snprintf(tmp, sizeof(tmp), "%d", h12); - } else if (ch == 'm') { - if (run >= 2) snprintf(tmp, sizeof(tmp), "%02d", tm.tm_min); - else snprintf(tmp, sizeof(tmp), "%d", tm.tm_min); - } else if (ch == 's') { - if (run >= 2) snprintf(tmp, sizeof(tmp), "%02d", tm.tm_sec); - else snprintf(tmp, sizeof(tmp), "%d", tm.tm_sec); - } else if (ch == 'a') { - snprintf(tmp, sizeof(tmp), "%s", tm.tm_hour < 12 ? "AM" : "PM"); - } else if (ch == 'E') { - snprintf(tmp, sizeof(tmp), "%s", _el_weekday_short[tm.tm_wday]); - } else if (ch == 'z') { - snprintf(tmp, sizeof(tmp), "%s", abbr); - } else { - for (size_t k = 0; k < run; k++) { - if (pos + 1 >= cap) { cap *= 2; out = realloc(out, cap); } - out[pos++] = ch; - } - i += run; - continue; - } - size_t tl = strlen(tmp); - if (pos + tl + 1 >= cap) { cap = (cap + tl) * 2; out = realloc(out, cap); } - memcpy(out + pos, tmp, tl); - pos += tl; - i += run; - } - out[pos] = '\0'; - char* result = el_strdup(out); - free(out); - return result; -} - -/* Format a Mars CalendarTime: %sol prints the integer sol number since - * mission epoch (Unix epoch fallback), %phase prints cycle_phase as a - * 0..1 decimal. Other %-specifiers fall through. */ -static char* _el_format_mars(el_caltime_t* ct, const char* pattern) { - el_calendar_t* c = ct->cal; - int64_t period = c->cycle_period_ns > 0 ? c->cycle_period_ns : EL_MARS_SOL_NS; - int64_t base = ct->instant_ns - c->epoch_ns; - int64_t sol = base / period; - int64_t phase_ns = base % period; - if (phase_ns < 0) { phase_ns += period; sol -= 1; } - double phase = (double)phase_ns / (double)period; - size_t cap = strlen(pattern) * 4 + 64; - char* out = (char*)malloc(cap); - size_t pos = 0; - for (size_t i = 0; pattern[i]; i++) { - if (pattern[i] == '%' && pattern[i+1]) { - char tmp[64]; - tmp[0] = '\0'; - if (strncmp(pattern + i + 1, "sol", 3) == 0) { - snprintf(tmp, sizeof(tmp), "%lld", (long long)sol); - i += 3; - } else if (strncmp(pattern + i + 1, "phase", 5) == 0) { - snprintf(tmp, sizeof(tmp), "%.4f", phase); - i += 5; - } else if (pattern[i+1] == 'd') { - snprintf(tmp, sizeof(tmp), "%lld", (long long)sol); - i += 1; - } else { - tmp[0] = pattern[i+1]; tmp[1] = '\0'; - i += 1; - } - size_t tl = strlen(tmp); - if (pos + tl + 1 >= cap) { cap = (cap + tl) * 2; out = realloc(out, cap); } - memcpy(out + pos, tmp, tl); - pos += tl; - } else { - if (pos + 1 >= cap) { cap *= 2; out = realloc(out, cap); } - out[pos++] = pattern[i]; - } - } - out[pos] = '\0'; - char* result = el_strdup(out); - free(out); - return result; -} - -/* Format a CycleCalendar CalendarTime: %cycle and %phase. */ -static char* _el_format_cycle(el_caltime_t* ct, const char* pattern) { - el_calendar_t* c = ct->cal; - int64_t period = c->cycle_period_ns > 0 ? c->cycle_period_ns : 1; - int64_t base = ct->instant_ns - c->epoch_ns; - int64_t cycle = base / period; - int64_t phase_ns = base % period; - if (phase_ns < 0) { phase_ns += period; cycle -= 1; } - double phase = (double)phase_ns / (double)period; - size_t cap = strlen(pattern) * 4 + 64; - char* out = (char*)malloc(cap); - size_t pos = 0; - for (size_t i = 0; pattern[i]; i++) { - if (pattern[i] == '%' && pattern[i+1]) { - char tmp[64]; - tmp[0] = '\0'; - if (strncmp(pattern + i + 1, "cycle", 5) == 0) { - snprintf(tmp, sizeof(tmp), "%lld", (long long)cycle); - i += 5; - } else if (strncmp(pattern + i + 1, "phase", 5) == 0) { - snprintf(tmp, sizeof(tmp), "%.4f", phase); - i += 5; - } else if (pattern[i+1] == 'd') { - snprintf(tmp, sizeof(tmp), "%lld", (long long)cycle); - i += 1; - } else if (pattern[i+1] == 'f') { - snprintf(tmp, sizeof(tmp), "%.2f", phase); - i += 1; - } else { - /* Pass through unknown specifier */ - tmp[0] = '%'; tmp[1] = pattern[i+1]; tmp[2] = '\0'; - i += 1; - } - size_t tl = strlen(tmp); - if (pos + tl + 1 >= cap) { cap = (cap + tl) * 2; out = realloc(out, cap); } - memcpy(out + pos, tmp, tl); - pos += tl; - } else { - if (pos + 1 >= cap) { cap *= 2; out = realloc(out, cap); } - out[pos++] = pattern[i]; - } - } - out[pos] = '\0'; - char* result = el_strdup(out); - free(out); - return result; -} - -el_val_t cal_format(el_val_t ct_val, el_val_t pattern_val) { - if (!el_is_magic(ct_val, EL_CALTIME_MAGIC)) return el_wrap_str(el_strdup("")); - el_caltime_t* ct = (el_caltime_t*)(uintptr_t)ct_val; - const char* pat = EL_CSTR(pattern_val); - if (!pat) pat = ""; - char* result = NULL; - switch (ct->cal->kind) { - case EL_CALENDAR_EARTH: result = _el_format_earth(ct, pat); break; - case EL_CALENDAR_MARS: result = _el_format_mars(ct, pat); break; - case EL_CALENDAR_CYCLE: result = _el_format_cycle(ct, pat); break; - case EL_CALENDAR_RELATIVE: result = _el_format_cycle(ct, pat); break; - case EL_CALENDAR_NO_CYCLE: { - char buf[64]; - snprintf(buf, sizeof(buf), "instant:%lld", (long long)ct->instant_ns); - result = el_strdup(buf); - break; - } - default: result = el_strdup(""); - } - return el_wrap_str(result); -} - -/* ── LocalDate / LocalTime / LocalDateTime ──────────────────────────────── */ - -static int _el_days_in_month(int y, int m) { - static const int dim[12] = {31,28,31,30,31,30,31,31,30,31,30,31}; - if (m == 2) { - int leap = ((y % 4 == 0) && (y % 100 != 0)) || (y % 400 == 0); - return 28 + (leap ? 1 : 0); - } - if (m < 1 || m > 12) return 30; - return dim[m - 1]; -} - -el_val_t local_date(el_val_t y, el_val_t m, el_val_t d) { - el_localdate_t* ld = (el_localdate_t*)malloc(sizeof(el_localdate_t)); - ld->magic = EL_LDATE_MAGIC; - ld->year = (int)(int64_t)y; - ld->month = (int)(int64_t)m; - ld->day = (int)(int64_t)d; - return (el_val_t)(uintptr_t)ld; -} - -el_val_t local_time(el_val_t h, el_val_t m, el_val_t s, el_val_t ns) { - int64_t hh = (int64_t)h; - int64_t mm = (int64_t)m; - int64_t ss = (int64_t)s; - int64_t nn = (int64_t)ns; - int64_t total = hh * 3600000000000LL + mm * 60000000000LL + ss * 1000000000LL + nn; - return (el_val_t)total; -} - -el_val_t local_datetime(el_val_t date_val, el_val_t time_val) { - if (!el_is_magic(date_val, EL_LDATE_MAGIC)) return (el_val_t)0; - el_localdt_t* ldt = (el_localdt_t*)malloc(sizeof(el_localdt_t)); - ldt->magic = EL_LDT_MAGIC; - ldt->date = (el_localdate_t*)(uintptr_t)date_val; - ldt->time_ns = (int64_t)time_val; - return (el_val_t)(uintptr_t)ldt; -} - -el_val_t zoned(el_val_t date_val, el_val_t time_val, el_val_t cal_val) { - if (!el_is_magic(date_val, EL_LDATE_MAGIC)) return (el_val_t)0; - el_localdate_t* ld = (el_localdate_t*)(uintptr_t)date_val; - el_calendar_t* c = _el_resolve_cal(cal_val); - int64_t time_ns = (int64_t)time_val; - /* Convert (LocalDate, LocalTime, EarthCalendar) -> Instant. - * For non-Earth calendars we use day-anchored conversion: treat the - * LocalDate's (y,m,d) as a Gregorian projection, convert to seconds via - * mktime under the calendar's zone, then add nanos-since-midnight. */ - if (c->kind == EL_CALENDAR_EARTH) { - _el_apply_zone(c->zone); - struct tm tm; memset(&tm, 0, sizeof(tm)); - tm.tm_year = ld->year - 1900; - tm.tm_mon = ld->month - 1; - tm.tm_mday = ld->day; - tm.tm_hour = (int)(time_ns / 3600000000000LL); - tm.tm_min = (int)((time_ns / 60000000000LL) % 60); - tm.tm_sec = (int)((time_ns / 1000000000LL) % 60); - tm.tm_isdst = -1; - time_t t = mktime(&tm); - if (t == (time_t)-1) return (el_val_t)0; - int64_t ns = (int64_t)t * 1000000000LL + (time_ns % 1000000000LL); - return (el_val_t)(uintptr_t)_el_caltime_alloc(ns, c); - } - /* Non-Earth fallback: project as if Earth UTC then attach calendar. */ - struct tm tm; memset(&tm, 0, sizeof(tm)); - tm.tm_year = ld->year - 1900; - tm.tm_mon = ld->month - 1; - tm.tm_mday = ld->day; - tm.tm_hour = (int)(time_ns / 3600000000000LL); - tm.tm_min = (int)((time_ns / 60000000000LL) % 60); - tm.tm_sec = (int)((time_ns / 1000000000LL) % 60); - time_t t = timegm(&tm); - if (t == (time_t)-1) return (el_val_t)0; - int64_t ns = (int64_t)t * 1000000000LL + (time_ns % 1000000000LL); - return (el_val_t)(uintptr_t)_el_caltime_alloc(ns, c); -} - -el_val_t local_date_year(el_val_t v) { - if (!el_is_magic(v, EL_LDATE_MAGIC)) return (el_val_t)0; - return (el_val_t)((el_localdate_t*)(uintptr_t)v)->year; -} -el_val_t local_date_month(el_val_t v) { - if (!el_is_magic(v, EL_LDATE_MAGIC)) return (el_val_t)0; - return (el_val_t)((el_localdate_t*)(uintptr_t)v)->month; -} -el_val_t local_date_day(el_val_t v) { - if (!el_is_magic(v, EL_LDATE_MAGIC)) return (el_val_t)0; - return (el_val_t)((el_localdate_t*)(uintptr_t)v)->day; -} -el_val_t local_time_hour(el_val_t v) { - int64_t t = (int64_t)v; - return (el_val_t)(t / 3600000000000LL); -} -el_val_t local_time_minute(el_val_t v) { - int64_t t = (int64_t)v; - return (el_val_t)((t / 60000000000LL) % 60); -} -el_val_t local_time_second(el_val_t v) { - int64_t t = (int64_t)v; - return (el_val_t)((t / 1000000000LL) % 60); -} -el_val_t local_time_nanos(el_val_t v) { - int64_t t = (int64_t)v; - return (el_val_t)(t % 1000000000LL); -} - -el_val_t el_local_date_add_dur(el_val_t ld_val, el_val_t dur_val) { - if (!el_is_magic(ld_val, EL_LDATE_MAGIC)) return ld_val; - el_localdate_t* ld = (el_localdate_t*)(uintptr_t)ld_val; - int64_t dur_ns = (int64_t)dur_val; - int64_t days = dur_ns / EL_EARTH_DAY_NS; - int y = ld->year, m = ld->month, d = ld->day; - /* Walk days forward/backward in canonical Gregorian. */ - while (days > 0) { - int dim = _el_days_in_month(y, m); - if (d + days <= dim) { d += (int)days; days = 0; break; } - days -= (dim - d + 1); - d = 1; - m++; - if (m > 12) { m = 1; y++; } - } - while (days < 0) { - if (d + days >= 1) { d += (int)days; days = 0; break; } - days += d; - m--; - if (m < 1) { m = 12; y--; } - d = _el_days_in_month(y, m); - } - return local_date((el_val_t)y, (el_val_t)m, (el_val_t)d); -} - -el_val_t el_local_time_add_dur(el_val_t lt_val, el_val_t dur_val) { - int64_t t = (int64_t)lt_val + (int64_t)dur_val; - /* Wrap mod 24h on Earth-default. CycleCalendar wrapping requires the - * caller to use cal_in / cal_format for the right modulus. */ - int64_t day = EL_EARTH_DAY_NS; - int64_t r = t % day; - if (r < 0) r += day; - return (el_val_t)r; -} - -el_val_t el_local_date_lt(el_val_t a_val, el_val_t b_val) { - if (!el_is_magic(a_val, EL_LDATE_MAGIC) || !el_is_magic(b_val, EL_LDATE_MAGIC)) return (el_val_t)0; - el_localdate_t* a = (el_localdate_t*)(uintptr_t)a_val; - el_localdate_t* b = (el_localdate_t*)(uintptr_t)b_val; - if (a->year != b->year) return (el_val_t)(a->year < b->year ? 1 : 0); - if (a->month != b->month) return (el_val_t)(a->month < b->month ? 1 : 0); - return (el_val_t)(a->day < b->day ? 1 : 0); -} - -el_val_t el_local_date_eq(el_val_t a_val, el_val_t b_val) { - if (!el_is_magic(a_val, EL_LDATE_MAGIC) || !el_is_magic(b_val, EL_LDATE_MAGIC)) return (el_val_t)0; - el_localdate_t* a = (el_localdate_t*)(uintptr_t)a_val; - el_localdate_t* b = (el_localdate_t*)(uintptr_t)b_val; - return (el_val_t)((a->year == b->year && a->month == b->month && a->day == b->day) ? 1 : 0); -} - -/* ── Rhythm ──────────────────────────────────────────────────────────────── */ - -static el_rhythm_t* _el_rhythm_alloc(el_rhythm_kind_t k) { - el_rhythm_t* r = (el_rhythm_t*)calloc(1, sizeof(el_rhythm_t)); - r->magic = EL_RHYTHM_MAGIC; - r->kind = k; - return r; -} - -el_val_t rhythm_cycle_start(void) { - return (el_val_t)(uintptr_t)_el_rhythm_alloc(EL_RHYTHM_CYCLE_START); -} - -el_val_t rhythm_cycle_phase(el_val_t phase_val) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_CYCLE_PHASE); - r->phase = el_to_float(phase_val); - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_duration(el_val_t d_val) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_DURATION); - r->period_ns = (int64_t)d_val; - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_session_start(void) { - return (el_val_t)(uintptr_t)_el_rhythm_alloc(EL_RHYTHM_SESSION_START); -} - -el_val_t rhythm_event(el_val_t name_val) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_EVENT); - const char* n = EL_CSTR(name_val); - r->event_name = el_strdup_persist(n ? n : ""); - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_and(el_val_t a_val, el_val_t b_val) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_AND); - r->a = el_is_magic(a_val, EL_RHYTHM_MAGIC) ? (el_rhythm_t*)(uintptr_t)a_val : NULL; - r->b = el_is_magic(b_val, EL_RHYTHM_MAGIC) ? (el_rhythm_t*)(uintptr_t)b_val : NULL; - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_or(el_val_t a_val, el_val_t b_val) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_OR); - r->a = el_is_magic(a_val, EL_RHYTHM_MAGIC) ? (el_rhythm_t*)(uintptr_t)a_val : NULL; - r->b = el_is_magic(b_val, EL_RHYTHM_MAGIC) ? (el_rhythm_t*)(uintptr_t)b_val : NULL; - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_weekday(el_val_t day) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_WEEKDAY); - r->weekday = (int)(int64_t)day; - return (el_val_t)(uintptr_t)r; -} - -el_val_t rhythm_weekly_at(el_val_t day, el_val_t hour, el_val_t minute) { - el_rhythm_t* r = _el_rhythm_alloc(EL_RHYTHM_WEEKLY_AT); - r->weekday = (int)(int64_t)day; - r->hour = (int)(int64_t)hour; - r->minute = (int)(int64_t)minute; - return (el_val_t)(uintptr_t)r; -} - -/* Compute the next instant on or after `after` when rhythm `r` matches, - * under calendar `cal`. */ -static int64_t _el_next_after(el_rhythm_t* r, int64_t after_ns, el_calendar_t* cal) { - if (!r) return after_ns; - int64_t period = cal->cycle_period_ns > 0 ? cal->cycle_period_ns : EL_EARTH_DAY_NS; - switch (r->kind) { - case EL_RHYTHM_CYCLE_START: { - int64_t base = after_ns - cal->epoch_ns; - int64_t cyc = (base / period) + 1; - return cal->epoch_ns + cyc * period; - } - case EL_RHYTHM_CYCLE_PHASE: { - int64_t base = after_ns - cal->epoch_ns; - int64_t cyc_ns = (int64_t)(r->phase * (double)period); - int64_t cur_cyc = base / period; - int64_t candidate = cal->epoch_ns + cur_cyc * period + cyc_ns; - if (candidate <= after_ns) candidate += period; - return candidate; - } - case EL_RHYTHM_DURATION: { - return after_ns + (r->period_ns > 0 ? r->period_ns : 1); - } - case EL_RHYTHM_WEEKDAY: - case EL_RHYTHM_WEEKLY_AT: { - if (cal->kind != EL_CALENDAR_EARTH) { - /* Non-Earth calendars: fall back to cycle math, treating - * weekday as a 7-cycle-per-period proxy. */ - return after_ns + period; - } - _el_apply_zone(cal->zone); - time_t s = (time_t)(after_ns / 1000000000LL); - struct tm tm; - localtime_r(&s, &tm); - /* tm_wday: 0=Sun..6=Sat. We use 1=Mon..7=Sun. */ - int target = r->weekday >= 1 && r->weekday <= 7 ? r->weekday : 1; - int target_wday = target == 7 ? 0 : target; /* 7→Sun=0, 1→Mon=1 */ - int days_ahead = (target_wday - tm.tm_wday + 7) % 7; - int hour = (r->kind == EL_RHYTHM_WEEKLY_AT) ? r->hour : 0; - int minute = (r->kind == EL_RHYTHM_WEEKLY_AT) ? r->minute : 0; - struct tm cand = tm; - cand.tm_mday += days_ahead; - cand.tm_hour = hour; - cand.tm_min = minute; - cand.tm_sec = 0; - cand.tm_isdst = -1; - time_t cand_t = mktime(&cand); - int64_t cand_ns = (int64_t)cand_t * 1000000000LL; - if (cand_ns <= after_ns) { - cand.tm_mday += 7; - cand.tm_isdst = -1; - cand_t = mktime(&cand); - cand_ns = (int64_t)cand_t * 1000000000LL; - } - return cand_ns; - } - case EL_RHYTHM_AND: { - int64_t a = _el_next_after(r->a, after_ns, cal); - int64_t b = _el_next_after(r->b, after_ns, cal); - return a > b ? a : b; - } - case EL_RHYTHM_OR: { - int64_t a = _el_next_after(r->a, after_ns, cal); - int64_t b = _el_next_after(r->b, after_ns, cal); - return a < b ? a : b; - } - case EL_RHYTHM_SESSION_START: - case EL_RHYTHM_EVENT: - default: - return after_ns; - } -} - -el_val_t rhythm_next_after(el_val_t r_val, el_val_t after_val, el_val_t cal_val) { - if (!el_is_magic(r_val, EL_RHYTHM_MAGIC)) return after_val; - el_rhythm_t* r = (el_rhythm_t*)(uintptr_t)r_val; - el_calendar_t* c = _el_resolve_cal(cal_val); - int64_t out = _el_next_after(r, (int64_t)after_val, c); - return (el_val_t)out; -} - -el_val_t rhythm_matches(el_val_t r_val, el_val_t ct_val) { - if (!el_is_magic(r_val, EL_RHYTHM_MAGIC)) return (el_val_t)0; - if (!el_is_magic(ct_val, EL_CALTIME_MAGIC)) return (el_val_t)0; - el_rhythm_t* r = (el_rhythm_t*)(uintptr_t)r_val; - el_caltime_t* ct = (el_caltime_t*)(uintptr_t)ct_val; - int64_t period = ct->cal->cycle_period_ns > 0 ? ct->cal->cycle_period_ns : EL_EARTH_DAY_NS; - int64_t base = ct->instant_ns - ct->cal->epoch_ns; - int64_t phase_ns = base % period; - if (phase_ns < 0) phase_ns += period; - double phase = (double)phase_ns / (double)period; - switch (r->kind) { - case EL_RHYTHM_CYCLE_START: return (el_val_t)(phase_ns == 0 ? 1 : 0); - case EL_RHYTHM_CYCLE_PHASE: { - double diff = phase - r->phase; - if (diff < 0) diff = -diff; - return (el_val_t)(diff < 0.001 ? 1 : 0); - } - default: return (el_val_t)0; - } -} - -/* ── UUID v4 ─────────────────────────────────────────────────────────────── */ - -static int _el_uuid_seeded = 0; - -static void _el_uuid_seed(void) { - if (!_el_uuid_seeded) { - srand((unsigned)time(NULL) ^ (unsigned)(uintptr_t)&_el_uuid_seeded); - _el_uuid_seeded = 1; - } -} - -el_val_t uuid_new(void) { - _el_uuid_seed(); - unsigned char b[16]; - for (int i = 0; i < 16; i++) b[i] = (unsigned char)(rand() & 0xff); - /* Version 4 */ - b[6] = (b[6] & 0x0f) | 0x40; - /* RFC 4122 variant */ - b[8] = (b[8] & 0x3f) | 0x80; - char buf[37]; - snprintf(buf, sizeof(buf), - "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x", - b[0], b[1], b[2], b[3], - b[4], b[5], - b[6], b[7], - b[8], b[9], - b[10], b[11], b[12], b[13], b[14], b[15]); - return el_wrap_str(el_strdup(buf)); -} - -el_val_t uuid_v4(void) { return uuid_new(); } - -/* ── Environment ─────────────────────────────────────────────────────────── */ - -el_val_t env(el_val_t key) { - const char* k = EL_CSTR(key); - if (!k) return el_wrap_str(el_strdup("")); - const char* v = getenv(k); - return el_wrap_str(el_strdup(v ? v : "")); -} - -/* ── In-process state K/V ────────────────────────────────────────────────── */ - -typedef struct { - char* key; - char* value; -} StateEntry; - -static StateEntry* _state_entries = NULL; -static size_t _state_count = 0; -static size_t _state_cap = 0; -/* Mutex protecting all _state_entries access. state_set/state_get are called - * concurrently from 64 HTTP worker threads — without this lock, realloc and - * free race, producing corruption, double-free, and segfaults. */ -static pthread_mutex_t _state_mu = PTHREAD_MUTEX_INITIALIZER; - -static StateEntry* state_find(const char* key) { - for (size_t i = 0; i < _state_count; i++) { - if (strcmp(_state_entries[i].key, key) == 0) return &_state_entries[i]; - } - return NULL; -} - -el_val_t state_set(el_val_t key, el_val_t value) { - const char* k = EL_CSTR(key); - const char* v = EL_CSTR(value); - if (!k) return 0; - if (!v) v = ""; - pthread_mutex_lock(&_state_mu); - StateEntry* e = state_find(k); - if (e) { - free(e->value); - e->value = el_strdup_persist(v); - pthread_mutex_unlock(&_state_mu); - return 1; - } - if (_state_count >= _state_cap) { - size_t nc = _state_cap == 0 ? 16 : _state_cap * 2; - StateEntry* grown = realloc(_state_entries, nc * sizeof(StateEntry)); - if (!grown) { pthread_mutex_unlock(&_state_mu); fputs("el_runtime: out of memory\n", stderr); exit(1); } - _state_entries = grown; - _state_cap = nc; - } - _state_entries[_state_count].key = el_strdup_persist(k); - _state_entries[_state_count].value = el_strdup_persist(v); - _state_count++; - pthread_mutex_unlock(&_state_mu); - return 1; -} - -el_val_t state_get(el_val_t key) { - const char* k = EL_CSTR(key); - if (!k) return el_wrap_str(el_strdup("")); - pthread_mutex_lock(&_state_mu); - StateEntry* e = state_find(k); - char* result = el_strdup_persist(e ? e->value : ""); - pthread_mutex_unlock(&_state_mu); - /* wrap in arena-tracked copy for the caller's request lifetime */ - char* copy = el_strdup(result); - return el_wrap_str(copy); -} - -el_val_t state_del(el_val_t key) { - const char* k = EL_CSTR(key); - if (!k) return 0; - pthread_mutex_lock(&_state_mu); - for (size_t i = 0; i < _state_count; i++) { - if (strcmp(_state_entries[i].key, k) == 0) { - free(_state_entries[i].key); - free(_state_entries[i].value); - for (size_t j = i + 1; j < _state_count; j++) { - _state_entries[j - 1] = _state_entries[j]; - } - _state_count--; - pthread_mutex_unlock(&_state_mu); - return 1; - } - } - pthread_mutex_unlock(&_state_mu); - return 1; -} - -el_val_t state_keys(void) { - pthread_mutex_lock(&_state_mu); - el_val_t lst = el_list_empty(); - for (size_t i = 0; i < _state_count; i++) { - lst = el_list_append(lst, el_wrap_str(el_strdup(_state_entries[i].key))); - } - pthread_mutex_unlock(&_state_mu); - return lst; -} - -/* ── Float formatting ────────────────────────────────────────────────────── */ - -el_val_t float_to_str(el_val_t f) { - char buf[64]; - snprintf(buf, sizeof(buf), "%g", el_to_float(f)); - return el_wrap_str(el_strdup(buf)); -} - -el_val_t int_to_float(el_val_t n) { - return el_from_float((double)(int64_t)n); -} - -el_val_t float_to_int(el_val_t f) { - return (el_val_t)(int64_t)el_to_float(f); -} - -el_val_t format_float(el_val_t f, el_val_t decimals) { - int d = (int)(int64_t)decimals; - if (d < 0) d = 0; - if (d > 30) d = 30; - char buf[128]; - snprintf(buf, sizeof(buf), "%.*f", d, el_to_float(f)); - return el_wrap_str(el_strdup(buf)); -} - -el_val_t decimal_round(el_val_t f, el_val_t decimals) { - int d = (int)(int64_t)decimals; - if (d < 0) d = 0; - if (d > 15) d = 15; - double mul = pow(10.0, (double)d); - double v = el_to_float(f); - double r = (v >= 0.0 ? floor(v * mul + 0.5) : -floor(-v * mul + 0.5)) / mul; - return el_from_float(r); -} - -el_val_t str_to_float(el_val_t s) { - const char* str = EL_CSTR(s); - if (!str) return el_from_float(0.0); - return el_from_float(strtod(str, NULL)); -} - -/* ── Math (Float-aware) ──────────────────────────────────────────────────── */ - -el_val_t math_sqrt(el_val_t f) { return el_from_float(sqrt(el_to_float(f))); } -el_val_t math_log(el_val_t f) { return el_from_float(log(el_to_float(f))); } -el_val_t math_ln(el_val_t f) { return el_from_float(log(el_to_float(f))); } -el_val_t math_sin(el_val_t f) { return el_from_float(sin(el_to_float(f))); } -el_val_t math_cos(el_val_t f) { return el_from_float(cos(el_to_float(f))); } -el_val_t math_pi(void) { return el_from_float(3.141592653589793238462643383279502884); } - -/* ── String additions ────────────────────────────────────────────────────── */ - -el_val_t str_index_of(el_val_t s, el_val_t sub) { - const char* str = EL_CSTR(s); - const char* sb = EL_CSTR(sub); - if (!str || !sb) return -1; - const char* hit = strstr(str, sb); - if (!hit) return -1; - return (el_val_t)(int64_t)(hit - str); -} - -el_val_t str_split(el_val_t s, el_val_t sep) { - const char* str = EL_CSTR(s); - const char* sp = EL_CSTR(sep); - el_val_t lst = el_list_empty(); - if (!str) return lst; - if (!sp || !*sp) { - lst = el_list_append(lst, el_wrap_str(el_strdup(str))); - return lst; - } - size_t lp = strlen(sp); - const char* p = str; - const char* hit; - while ((hit = strstr(p, sp)) != NULL) { - size_t n = (size_t)(hit - p); - char* out = el_strbuf(n); - memcpy(out, p, n); - out[n] = '\0'; - lst = el_list_append(lst, el_wrap_str(out)); - p = hit + lp; - } - lst = el_list_append(lst, el_wrap_str(el_strdup(p))); - return lst; -} - -el_val_t str_char_at(el_val_t s, el_val_t i) { - const char* str = EL_CSTR(s); - int64_t idx = (int64_t)i; - if (!str) return el_wrap_str(el_strdup("")); - int64_t n = (int64_t)strlen(str); - if (idx < 0 || idx >= n) return el_wrap_str(el_strdup("")); - char buf[2]; - buf[0] = str[idx]; - buf[1] = '\0'; - return el_wrap_str(el_strdup(buf)); -} - -el_val_t str_char_code(el_val_t s, el_val_t i) { - const char* str = EL_CSTR(s); - int64_t idx = (int64_t)i; - if (!str) return 0; - int64_t n = (int64_t)strlen(str); - if (idx < 0 || idx >= n) return 0; - return (el_val_t)(unsigned char)str[idx]; -} - -static el_val_t str_pad(const char* s, int64_t width, const char* pad, int left) { - if (!s) s = ""; - if (!pad || !*pad) pad = " "; - int64_t lp = (int64_t)strlen(pad); - int64_t ls = (int64_t)strlen(s); - if (ls >= width) return el_wrap_str(el_strdup(s)); - int64_t need = width - ls; - char* out = el_strbuf((size_t)width); - if (left) { - for (int64_t i = 0; i < need; i++) out[i] = pad[i % lp]; - memcpy(out + need, s, (size_t)ls); - } else { - memcpy(out, s, (size_t)ls); - for (int64_t i = 0; i < need; i++) out[ls + i] = pad[i % lp]; - } - out[width] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_pad_left(el_val_t s, el_val_t width, el_val_t pad) { - return str_pad(EL_CSTR(s), (int64_t)width, EL_CSTR(pad), 1); -} - -el_val_t str_pad_right(el_val_t s, el_val_t width, el_val_t pad) { - return str_pad(EL_CSTR(s), (int64_t)width, EL_CSTR(pad), 0); -} - -el_val_t str_format(el_val_t fmt, el_val_t data) { - const char* tpl = EL_CSTR(fmt); - if (!tpl) return el_wrap_str(el_strdup("")); - JsonBuf b; jb_init(&b); - const char* p = tpl; - while (*p) { - if (*p == '{') { - const char* q = p + 1; - while (*q && *q != '}') q++; - if (*q == '}') { - size_t klen = (size_t)(q - p - 1); - char keybuf[256]; - if (klen < sizeof(keybuf)) { - memcpy(keybuf, p + 1, klen); - keybuf[klen] = '\0'; - el_val_t v = el_map_get(data, EL_STR(keybuf)); - if (v != 0 && looks_like_string(v)) { - jb_puts(&b, EL_CSTR(v)); - p = q + 1; - continue; - } else if (v != 0) { - jb_emit_int(&b, (int64_t)v); - p = q + 1; - continue; - } - } - /* Unknown key — leave {key} verbatim */ - jb_reserve(&b, klen + 2); - memcpy(b.buf + b.len, p, klen + 2); - b.len += klen + 2; - b.buf[b.len] = '\0'; - p = q + 1; - continue; - } - } - jb_putc(&b, *p); - p++; - } - return el_wrap_str(b.buf); -} - -el_val_t str_lower(el_val_t s) { return str_to_lower(s); } -el_val_t str_upper(el_val_t s) { return str_to_upper(s); } - -/* ── Text-processing primitives (Phase 1: byte/codepoint, ASCII char classes) - * - * Phase 1 covers the operations every text-handling caller used to roll by - * hand on top of str_index_of + str_slice. The character-class predicates - * (is_letter / is_digit / ...) are ASCII only — Unicode-grapheme awareness, - * NFC/NFD normalization, and regex are Phase 2. Single-char input checks the - * first byte; multi-char input requires ALL bytes to match (false otherwise). - * - * Counting: - * str_count non-overlapping occurrences of sub in s - * str_count_chars codepoint count (UTF-8 leading-byte count) - * str_count_bytes explicit byte length (alias of str_len) - * str_count_lines \n-delimited line count (\r\n folded to \n) - * str_count_words whitespace-delimited tokens, non-empty only - * str_count_letters ASCII [A-Za-z] - * str_count_digits ASCII [0-9] - * - * Find / position: - * str_index_of_all all byte offsets of sub, [] if none - * str_last_index_of last byte offset of sub, -1 if not found - * str_find_chars first index of any char in any_of, -1 if none - * - * Transform: - * str_repeat s * n (non-negative) - * str_reverse codepoint-reversed (NOT grapheme-aware) - * str_strip_prefix s without prefix if present, else s - * str_strip_suffix s without suffix if present, else s - * str_strip_chars strip leading+trailing chars matching any in chars - * str_lstrip strip leading whitespace - * str_rstrip strip trailing whitespace - * - * Char classification (Bool): - * is_letter, is_digit, is_alphanumeric, is_whitespace, - * is_punctuation, is_uppercase, is_lowercase - * - * Splitting: - * str_split_lines \n-delimited (\r\n folded). Trailing empty dropped. - * str_split_chars alias of native_string_chars in str_ namespace - * str_split_n split into at most n parts (last part keeps the - * rest verbatim, including any further separators) - * - * Joining: - * str_join [String] -> String, sep between elements - */ - -/* Count non-overlapping occurrences of sub in s. Empty sub returns 0. */ -el_val_t str_count(el_val_t sv, el_val_t subv) { - const char* s = EL_CSTR(sv); - const char* sub = EL_CSTR(subv); - if (!s || !sub || !*sub) return 0; - size_t lp = strlen(sub); - int64_t count = 0; - const char* p = s; - while ((p = strstr(p, sub)) != NULL) { - count++; - p += lp; /* non-overlapping advance */ - } - return (el_val_t)count; -} - -/* Codepoint count: walk bytes, count those NOT matching 10xxxxxx. */ -el_val_t str_count_chars(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - int64_t count = 0; - for (const unsigned char* p = (const unsigned char*)s; *p; p++) { - if ((*p & 0xC0) != 0x80) count++; - } - return (el_val_t)count; -} - -el_val_t str_count_bytes(el_val_t sv) { - return str_len(sv); -} - -el_val_t str_count_lines(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s || !*s) return 0; - int64_t count = 0; - int has_content = 0; - for (const char* p = s; *p; p++) { - has_content = 1; - if (*p == '\n') { - count++; - has_content = 0; /* the \n closed the line */ - } - } - if (has_content) count++; /* trailing line with no terminator */ - return (el_val_t)count; -} - -el_val_t str_count_words(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - int64_t count = 0; - int in_word = 0; - for (const unsigned char* p = (const unsigned char*)s; *p; p++) { - if (isspace(*p)) { - in_word = 0; - } else if (!in_word) { - in_word = 1; - count++; - } - } - return (el_val_t)count; -} - -el_val_t str_count_letters(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - int64_t count = 0; - for (const unsigned char* p = (const unsigned char*)s; *p; p++) { - if ((*p >= 'A' && *p <= 'Z') || (*p >= 'a' && *p <= 'z')) count++; - } - return (el_val_t)count; -} - -el_val_t str_count_digits(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return 0; - int64_t count = 0; - for (const unsigned char* p = (const unsigned char*)s; *p; p++) { - if (*p >= '0' && *p <= '9') count++; - } - return (el_val_t)count; -} - -el_val_t str_index_of_all(el_val_t sv, el_val_t subv) { - const char* s = EL_CSTR(sv); - const char* sub = EL_CSTR(subv); - el_val_t lst = el_list_empty(); - if (!s || !sub || !*sub) return lst; - size_t lp = strlen(sub); - const char* p = s; - const char* hit; - while ((hit = strstr(p, sub)) != NULL) { - lst = el_list_append(lst, (el_val_t)(int64_t)(hit - s)); - p = hit + lp; - } - return lst; -} - -el_val_t str_last_index_of(el_val_t sv, el_val_t subv) { - const char* s = EL_CSTR(sv); - const char* sub = EL_CSTR(subv); - if (!s || !sub || !*sub) return -1; - size_t lp = strlen(sub); - int64_t last = -1; - const char* p = s; - const char* hit; - while ((hit = strstr(p, sub)) != NULL) { - last = (int64_t)(hit - s); - p = hit + lp; - } - return (el_val_t)last; -} - -el_val_t str_find_chars(el_val_t sv, el_val_t any_of_v) { - const char* s = EL_CSTR(sv); - const char* any = EL_CSTR(any_of_v); - if (!s || !any || !*any) return -1; - for (const char* p = s; *p; p++) { - if (strchr(any, *p)) return (el_val_t)(int64_t)(p - s); - } - return -1; -} - -el_val_t str_repeat(el_val_t sv, el_val_t nv) { - const char* s = EL_CSTR(sv); - int64_t n = (int64_t)nv; - if (!s || n <= 0) return el_wrap_str(el_strdup("")); - size_t ls = strlen(s); - if (ls == 0) return el_wrap_str(el_strdup("")); - size_t total = ls * (size_t)n; - char* out = el_strbuf(total); - for (int64_t i = 0; i < n; i++) { - memcpy(out + i * ls, s, ls); - } - out[total] = '\0'; - return el_wrap_str(out); -} - -/* Reverse by codepoint: walk codepoints, copy each backwards into the output. - * NOT grapheme-aware (Phase 2). Combining marks attached to a base codepoint - * will detach. ASCII strings are byte-reverse equivalent. */ -el_val_t str_reverse(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - size_t n = strlen(s); - char* out = el_strbuf(n); - /* Walk forward, find each codepoint's byte length, then copy from the end. */ - size_t out_pos = n; - const unsigned char* p = (const unsigned char*)s; - while (*p) { - int cp_len; - if ((*p & 0x80) == 0x00) cp_len = 1; - else if ((*p & 0xE0) == 0xC0) cp_len = 2; - else if ((*p & 0xF0) == 0xE0) cp_len = 3; - else if ((*p & 0xF8) == 0xF0) cp_len = 4; - else cp_len = 1; /* invalid byte: passthrough */ - out_pos -= cp_len; - memcpy(out + out_pos, p, cp_len); - p += cp_len; - } - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_strip_prefix(el_val_t sv, el_val_t prefv) { - const char* s = EL_CSTR(sv); - const char* pref = EL_CSTR(prefv); - if (!s) return el_wrap_str(el_strdup("")); - if (!pref || !*pref) return el_wrap_str(el_strdup(s)); - size_t lp = strlen(pref); - size_t ls = strlen(s); - if (lp <= ls && strncmp(s, pref, lp) == 0) { - char* out = el_strbuf(ls - lp); - memcpy(out, s + lp, ls - lp); - out[ls - lp] = '\0'; - return el_wrap_str(out); - } - return el_wrap_str(el_strdup(s)); -} - -el_val_t str_strip_suffix(el_val_t sv, el_val_t sufv) { - const char* s = EL_CSTR(sv); - const char* suf = EL_CSTR(sufv); - if (!s) return el_wrap_str(el_strdup("")); - if (!suf || !*suf) return el_wrap_str(el_strdup(s)); - size_t ls = strlen(s); - size_t lsuf = strlen(suf); - if (lsuf <= ls && strcmp(s + ls - lsuf, suf) == 0) { - char* out = el_strbuf(ls - lsuf); - memcpy(out, s, ls - lsuf); - out[ls - lsuf] = '\0'; - return el_wrap_str(out); - } - return el_wrap_str(el_strdup(s)); -} - -el_val_t str_strip_chars(el_val_t sv, el_val_t charsv) { - const char* s = EL_CSTR(sv); - const char* chars = EL_CSTR(charsv); - if (!s) return el_wrap_str(el_strdup("")); - if (!chars || !*chars) return el_wrap_str(el_strdup(s)); - const char* start = s; - while (*start && strchr(chars, *start)) start++; - size_t n = strlen(start); - while (n > 0 && strchr(chars, start[n - 1])) n--; - char* out = el_strbuf(n); - memcpy(out, start, n); - out[n] = '\0'; - return el_wrap_str(out); -} - -el_val_t str_lstrip(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - while (*s && isspace((unsigned char)*s)) s++; - return el_wrap_str(el_strdup(s)); -} - -el_val_t str_rstrip(el_val_t sv) { - const char* s = EL_CSTR(sv); - if (!s) return el_wrap_str(el_strdup("")); - size_t n = strlen(s); - while (n > 0 && isspace((unsigned char)s[n - 1])) n--; - char* out = el_strbuf(n); - memcpy(out, s, n); - out[n] = '\0'; - return el_wrap_str(out); -} - -/* Character classification. - * Empty input returns false. Multi-char input requires ALL bytes to match. - * ASCII range only; Phase 2 will widen to Unicode. */ -static int s_all_match(el_val_t sv, int (*pred)(unsigned char)) { - const char* s = EL_CSTR(sv); - if (!s || !*s) return 0; - for (const unsigned char* p = (const unsigned char*)s; *p; p++) { - if (!pred(*p)) return 0; - } - return 1; -} - -static int p_letter(unsigned char c) { return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z'); } -static int p_digit(unsigned char c) { return c >= '0' && c <= '9'; } -static int p_alnum(unsigned char c) { return p_letter(c) || p_digit(c); } -static int p_white(unsigned char c) { return c == ' ' || c == '\t' || c == '\n' || c == '\r' || c == '\f' || c == '\v'; } -static int p_punct(unsigned char c) { return ispunct(c) ? 1 : 0; } -static int p_upper(unsigned char c) { return c >= 'A' && c <= 'Z'; } -static int p_lower(unsigned char c) { return c >= 'a' && c <= 'z'; } - -el_val_t is_letter(el_val_t s) { return (el_val_t)s_all_match(s, p_letter); } -el_val_t is_digit(el_val_t s) { return (el_val_t)s_all_match(s, p_digit); } -el_val_t is_alphanumeric(el_val_t s) { return (el_val_t)s_all_match(s, p_alnum); } -el_val_t is_whitespace(el_val_t s) { return (el_val_t)s_all_match(s, p_white); } -el_val_t is_punctuation(el_val_t s) { return (el_val_t)s_all_match(s, p_punct); } -el_val_t is_uppercase(el_val_t s) { return (el_val_t)s_all_match(s, p_upper); } -el_val_t is_lowercase(el_val_t s) { return (el_val_t)s_all_match(s, p_lower); } - -/* Split on \n. \r\n is folded to \n first. Trailing empty after final \n - * is dropped (so "a\nb\n" -> ["a", "b"], not ["a", "b", ""]). */ -el_val_t str_split_lines(el_val_t sv) { - const char* s = EL_CSTR(sv); - el_val_t lst = el_list_empty(); - if (!s) return lst; - size_t n = strlen(s); - /* Pre-scan: build into a normalized buffer with \r\n folded. */ - const char* line_start = s; - for (size_t i = 0; i <= n; i++) { - if (s[i] == '\n' || s[i] == '\0') { - size_t len = (size_t)(s + i - line_start); - /* Drop trailing \r if this was \r\n. */ - if (len > 0 && line_start[len - 1] == '\r') len--; - /* Drop final trailing-empty-after-newline. */ - if (s[i] == '\0' && len == 0 && i > 0 && s[i - 1] == '\n') break; - char* out = el_strbuf(len); - memcpy(out, line_start, len); - out[len] = '\0'; - lst = el_list_append(lst, el_wrap_str(out)); - if (s[i] == '\0') break; - line_start = s + i + 1; - } - } - return lst; -} - -el_val_t str_split_chars(el_val_t s) { - return native_string_chars(s); -} - -/* Split into at most n parts. The (n-1)th split point is the LAST split; - * after it, the remainder is appended verbatim including any further - * separators. n <= 0 returns an empty list. n == 1 returns [s]. */ -el_val_t str_split_n(el_val_t sv, el_val_t sepv, el_val_t nv) { - const char* s = EL_CSTR(sv); - const char* sep = EL_CSTR(sepv); - int64_t n = (int64_t)nv; - el_val_t lst = el_list_empty(); - if (!s) return lst; - if (n <= 0) return lst; - if (n == 1 || !sep || !*sep) { - lst = el_list_append(lst, el_wrap_str(el_strdup(s))); - return lst; - } - size_t lp = strlen(sep); - const char* p = s; - int64_t parts = 0; - const char* hit; - while (parts < n - 1 && (hit = strstr(p, sep)) != NULL) { - size_t len = (size_t)(hit - p); - char* out = el_strbuf(len); - memcpy(out, p, len); - out[len] = '\0'; - lst = el_list_append(lst, el_wrap_str(out)); - p = hit + lp; - parts++; - } - /* Remainder verbatim. */ - lst = el_list_append(lst, el_wrap_str(el_strdup(p))); - return lst; -} - -/* Join a [String] with a separator. Empty list -> "". Single-element -> - * that element. Non-string elements are stringified via int_to_str. */ -el_val_t str_join(el_val_t listv, el_val_t sepv) { - return list_join(listv, sepv); -} - -/* ── List additions ──────────────────────────────────────────────────────── */ - -el_val_t list_push(el_val_t list, el_val_t elem) { - return el_list_append(list, elem); -} - -el_val_t list_push_front(el_val_t listv, el_val_t elem) { - ElList* lst = (ElList*)(uintptr_t)listv; - if (!lst) { - el_val_t nl = el_list_empty(); - return el_list_append(nl, elem); - } - /* Append to grow capacity, then shift right */ - listv = el_list_append(listv, elem); - lst = (ElList*)(uintptr_t)listv; - for (int64_t i = lst->length - 1; i > 0; i--) { - lst->elems[i] = lst->elems[i - 1]; - } - lst->elems[0] = elem; - return EL_STR(lst); -} - -el_val_t list_join(el_val_t listv, el_val_t sep) { - ElList* lst = (ElList*)(uintptr_t)listv; - const char* sp = EL_CSTR(sep); - if (!sp) sp = ""; - if (!lst || lst->length == 0) return el_wrap_str(el_strdup("")); - JsonBuf b; jb_init(&b); - for (int64_t i = 0; i < lst->length; i++) { - if (i > 0) jb_puts(&b, sp); - el_val_t v = lst->elems[i]; - if (v == 0) continue; - if (looks_like_string(v)) { - jb_puts(&b, EL_CSTR(v)); - } else { - char tmp[32]; - snprintf(tmp, sizeof(tmp), "%lld", (long long)v); - jb_puts(&b, tmp); - } - } - return el_wrap_str(b.buf); -} - -el_val_t list_range(el_val_t start, el_val_t end) { - int64_t a = (int64_t)start; - int64_t b = (int64_t)end; - el_val_t lst = el_list_empty(); - for (int64_t i = a; i < b; i++) lst = el_list_append(lst, (el_val_t)i); - return lst; -} - -/* ── Bool helpers ────────────────────────────────────────────────────────── */ - -el_val_t bool_to_str(el_val_t b) { - return el_wrap_str(el_strdup(b ? "true" : "false")); -} - -/* ── Numeric parsing ─────────────────────────────────────────────────────── */ - -/* parse_int — strtoll with a default. str_to_int already exists but does not - * distinguish "0" from a parse failure, so callers that need a sentinel use - * this. Skips leading whitespace; accepts an optional leading +/-; returns - * default_val on empty input or no consumed digits. Trailing junk is ignored - * (atoi-style). */ -el_val_t parse_int(el_val_t sv, el_val_t default_val) { - const char* s = EL_CSTR(sv); - if (!s) return default_val; - while (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r') s++; - if (*s == '\0') return default_val; - char* end = NULL; - long long n = strtoll(s, &end, 10); - if (end == s) return default_val; - return (el_val_t)n; -} - -/* ── Process ─────────────────────────────────────────────────────────────── */ - -void exit_program(el_val_t code) { - exit((int)code); -} - -/* getpid_now — current process id. Named with the _now suffix to avoid - * colliding with the libc `getpid` declaration that the runtime already - * sees via (calling it `getpid` would fight the prototype). */ -el_val_t getpid_now(void) { - return (el_val_t)getpid(); -} - -/* ── args() — command-line argument access ────────────────────────────────── - * Compiled El programs call args() to get a list of CLI arguments. - * Call el_runtime_init_args(argc, argv) at the start of C main() to populate. - * The args list excludes argv[0] (the program name). */ - -static el_val_t _el_args_list = 0; - -void el_runtime_init_args(int argc, char** argv) { - _el_args_list = el_list_empty(); - for (int i = 1; i < argc; i++) { - _el_args_list = el_list_append(_el_args_list, EL_STR(argv[i])); - } -} - -el_val_t args(void) { - if (!_el_args_list) _el_args_list = el_list_empty(); - return _el_args_list; -} - -/* ── CGI identity ──────────────────────────────────────────────────────────── - * Called once at program start by the generated main() of a cgi {} program. - * Stores CGI identity so dharma_* builtins can reference it. */ - -static const char* _el_cgi_name = NULL; -static const char* _el_cgi_dharma_id = NULL; -static const char* _el_cgi_principal = NULL; -static const char* _el_cgi_network = NULL; -static const char* _el_cgi_engram = NULL; - -void el_cgi_init(el_val_t name, el_val_t dharma_id, el_val_t principal, - el_val_t network, el_val_t engram) { - _el_cgi_name = EL_CSTR(name); - _el_cgi_dharma_id = EL_CSTR(dharma_id); - _el_cgi_principal = EL_CSTR(principal); - _el_cgi_network = EL_CSTR(network) ? EL_CSTR(network) : "dharma-mainnet"; - _el_cgi_engram = EL_CSTR(engram) ? EL_CSTR(engram) : "http://localhost:8742"; - printf("[cgi] identity: name=%s dharma_id=%s principal=%s network=%s engram=%s\n", - _el_cgi_name ? _el_cgi_name : "(unset)", - _el_cgi_dharma_id ? _el_cgi_dharma_id : "(unset)", - _el_cgi_principal ? _el_cgi_principal : "(unset)", - _el_cgi_network, - _el_cgi_engram); -} - - -/* ── Batch 3: Engram in-process graph store ──────────────────────────────── */ -/* - * Single global EngramStore allocated lazily on first call. All node and - * edge content strings are owned (strdup'd) by the store. Linear arrays - * with doubling capacity for both nodes and edges. - * - * Two-layer activation algorithm (engram_activate): - * - * LAYER 1 — Broad fan-out (background activation): - * 1. Find seed nodes whose content/label/tags contain query (case-insens). - * 2. BFS up to `depth` hops along ALL edges (excitatory and inhibitory). - * Every reachable node fires — nothing is filtered at this layer. - * 3. bg_act = seed.salience * temporal_decay * dampening - * propagated as: new_bg = parent_bg * edge_weight * 0.7 * (1 + tbonus) - * where tbonus ∈ {0, 0.10, 0.20} for co-temporal nodes. - * 4. If reached by multiple paths, take max background_activation. - * 5. Persist background_activation to EngramNode.background_activation. - * - * LAYER 2 — Executive filter (working memory promotion): - * 6. For each inhibitory edge where source has background_activation > 0: - * inhibition[target] = max(bg[source] * e->weight) - * 7. For each background-activated node: - * raw_wm = bg * goal_bias(node, query) * confidence - * * (1 - (1 - INHIBITION_FACTOR) * inhibition) - * 8. Per-type threshold gate: raw_wm >= type_threshold → promoted. - * Safety/DharmaSelf: 0.05 Canonical: 0.15 Lesson: 0.25 - * Belief/Entity: 0.30 Note/Memory/Working: 0.40 - * 9. If not promoted: suppression_count++. After - * ENGRAM_SUPPRESSION_BREAKTHROUGH suppressions → force breakthrough - * at ENGRAM_BREAKTHROUGH_WEIGHT (latent tension surfacing). - * 10. Persist working_memory_weight to EngramNode.working_memory_weight. - * 11. Sort: promoted nodes (wm > 0) first by wm desc, then background- - * only by bg desc. Context compilation uses ONLY promoted nodes. - * - * Temporal decay: - * decay_factor = exp(-lambda * age_hours / T_half) - * T_half = 168.0 h (one week), lambda = ln(2) - * - * Activation dampening: - * dampen = 1.0 / (1.0 + log(1 + activation_count)) - * - * engram_query_range(start_ms, end_ms): - * Returns nodes whose created_at OR last_activated falls within - * [start_ms, end_ms], sorted by created_at ascending. - */ - -/* Temporal decay constants. - * T_HALF_HOURS: half-life in hours — one week. After one week of no - * activation a node retains 50% of its base salience contribution. - * DECAY_LAMBDA: ln(2) ≈ 0.693147 */ -#define ENGRAM_T_HALF_HOURS 168.0 -#define ENGRAM_DECAY_LAMBDA 0.693147 - -/* Two-layer activation constants. - * ENGRAM_WM_THRESHOLD: minimum background_activation for a node to be - * considered for working-memory promotion (layer 2 candidate gate). - * ENGRAM_WM_DECAY: per-turn decay applied to working_memory_weight for - * nodes NOT re-activated in the current turn (conversational thread - * continuity: a node promoted in turn N persists with reduced weight - * into turn N+1 without re-activation cost). - * ENGRAM_SUPPRESSION_BREAKTHROUGH: after this many consecutive suppressions - * a latent node forces itself into working memory at reduced weight, - * modelling the brain's "intrusive thought" / unresolved-tension surfacing. - * ENGRAM_BREAKTHROUGH_WEIGHT: the reduced working_memory_weight assigned - * when a suppressed node breaks through. - * ENGRAM_INHIBITION_FACTOR: multiplier applied to working_memory_weight when - * an inhibitory edge fires against a node (0 = full suppress, 0.3 = partial). */ -#define ENGRAM_WM_THRESHOLD 0.15 -#define ENGRAM_WM_DECAY 0.7 -#define ENGRAM_SUPPRESSION_BREAKTHROUGH 5 -#define ENGRAM_BREAKTHROUGH_WEIGHT 0.25 -#define ENGRAM_INHIBITION_FACTOR 0.1 - -/* ── Layered consciousness architecture ────────────────────────────────────── - * - * The engram graph is stratified into LAYERS that gate which suppressions - * apply during the executive filter pass. Layers are ordered shallow-to-deep - * by `activation_priority`; the deepest layer (priority 0, conventionally - * "safety") is the structural floor of the soul: nodes here cannot be - * silenced by inhibitory edges from any other layer. Higher layers - * (core-identity, domain-knowledge, imprint, suit) are normally - * suppressible — they participate in attentional inhibition and goal - * focus the way the prior single-graph implementation did. - * - * The five canonical layers (see engram_init_layers): - * 0. safety — structural, transparent, non-injectable, non-suppressible - * 1. core-identity — default for legacy nodes; suppressible - * 2. domain-knowledge— suppressible - * 3. imprint — runtime-injectable (an Imprint package can add/remove) - * 4. suit — runtime-injectable (a Suit overlays domain skill) - * - * Three-pass activation (engram_activate): - * Pass 1 — Background fan-out: BFS spreads activation across ALL layers - * (existing behavior preserved). Inhibitory edges propagate at - * this layer too; no filtering happens here. - * Pass 2 — Working memory promotion: type-threshold gate, goal bias, - * confidence weighting, inhibitory suppression. Inhibitory edges - * ONLY apply against nodes whose layer is `suppressible == 1`. - * Nodes in non-suppressible layers (Layer 0) ignore inhibition. - * Pass 3 — Layer 0 override: every node in a non-suppressible layer that - * received background activation has its working_memory_weight - * forced to >= ENGRAM_LAYER0_OVERRIDE_WEIGHT. The sacred fire — - * safety nodes that touched any seed unconditionally surface, - * even when the executive filter would have silenced them. - * - * Layer fields: - * suppressible : 0 → inhibitory edges are ignored against nodes in this - * layer during pass 2. Pass 3 also force-promotes them. - * 1 → standard behavior (most layers). - * transparent : 1 → emitted into the prompt context so its content shapes - * output, but filtered out of "what do you know about - * yourself?" introspection queries (engram_search and - * friends do not return transparent-layer nodes by - * default). 0 → fully visible to introspection. - * injectable : 1 → can be added/removed at runtime via engram_add_layer - * and engram_remove_layer (imprints, suits). - * 0 → built-in, fixed at engram_get() initialization. - * - * Backward compatibility: - * Nodes and edges loaded from snapshots without a `layer_id` field default - * to layer 1 (core-identity). The five canonical layers are always present. - */ -#define ENGRAM_LAYER_SAFETY 0u -#define ENGRAM_LAYER_CORE_IDENTITY 1u -#define ENGRAM_LAYER_DOMAIN 2u -#define ENGRAM_LAYER_IMPRINT 3u -#define ENGRAM_LAYER_SUIT 4u -#define ENGRAM_LAYER_DEFAULT ENGRAM_LAYER_CORE_IDENTITY - -/* Pass 3 override floor. Layer 0 nodes that received any background - * activation are force-promoted to AT LEAST this working_memory_weight, - * regardless of inhibitory suppression in pass 2. */ -#define ENGRAM_LAYER0_OVERRIDE_WEIGHT 1.0 - -/* Per-node-type activation thresholds. - * Lower tier / safety-critical nodes fire more readily. */ -static double engram_type_threshold(const char* node_type, const char* tier) { - if (node_type) { - if (strcmp(node_type, "DharmaSelf") == 0) return 0.05; - if (strcmp(node_type, "Safety") == 0) return 0.05; - } - if (tier) { - if (strcmp(tier, "Canonical") == 0) return 0.15; - if (strcmp(tier, "Lesson") == 0) return 0.25; - } - if (node_type) { - if (strcmp(node_type, "Belief") == 0) return 0.30; - if (strcmp(node_type, "Entity") == 0) return 0.30; - } - return 0.40; /* Note / Memory / Working (most nodes) */ -} - -typedef struct EngramNode { - char* id; - char* content; - char* node_type; - char* label; - char* tier; - char* tags; - char* metadata; - double salience; - double importance; - double confidence; - double temporal_decay_rate; /* per-node override for lambda; 0 = use default */ - int64_t activation_count; - int64_t last_activated; - int64_t created_at; - int64_t updated_at; - /* Two-layer activation fields ───────────────────────────────────────── - * background_activation: Layer 1. Set by BFS fan-out on every query. - * Every reachable node fires here — nothing is filtered at this stage. - * Models the brain's massive parallel sub-threshold activation of all - * associated content in response to a stimulus. - * working_memory_weight: Layer 2. Executive filter output. Only nodes - * that survive goal-state / attentional-bias scoring receive a - * non-zero weight here. Context compilation ONLY uses this field. - * Background-activated nodes with working_memory_weight == 0 remain - * latent — real, available, but silent. - * suppression_count: Consecutive turn count where this node was - * background-activated but NOT promoted to working memory. High - * values signal the node "wants to surface." After - * ENGRAM_SUPPRESSION_BREAKTHROUGH consecutive suppressions the node - * is force-promoted at a reduced weight (breakthrough activation). */ - double background_activation; - double working_memory_weight; - int32_t suppression_count; - /* Layered consciousness — see ENGRAM_LAYER_* macros and engram_init_layers. - * Defaults to ENGRAM_LAYER_DEFAULT (1, core-identity) for legacy nodes - * created via engram_node / engram_node_full and for snapshots that - * predate the layered schema. */ - uint32_t layer_id; -} EngramNode; - -typedef struct EngramEdge { - char* id; - char* from_id; - char* to_id; - char* relation; - char* metadata; - double weight; - double confidence; - int64_t created_at; - int64_t updated_at; - int64_t last_fired; - /* Inhibitory flag: when 1, activating the source node SUPPRESSES the - * working_memory_weight of the target node rather than exciting it. - * Models attentional inhibition: "I am focused on code work" creates - * inhibitory edges to personal/emotional nodes, preventing them from - * surfacing even if they have high background_activation. */ - int inhibitory; - /* Layered consciousness — edges carry a layer assignment for - * categorization/visualization. Pass 2 inhibitory gating is decided by - * the TARGET node's layer (whether it's suppressible), not by the edge - * layer. Defaults to ENGRAM_LAYER_DEFAULT. */ - uint32_t layer_id; -} EngramEdge; - -/* Layered consciousness — runtime layer registry entry. */ -typedef struct EngramLayer { - uint32_t layer_id; /* 0 = deepest (safety/limbic) */ - char* name; /* persistent — owned by the store */ - uint32_t activation_priority; /* lower = fires earlier; safety = 0 */ - int suppressible; /* can higher layers suppress nodes here? */ - int transparent; /* invisible to introspection queries? */ - int injectable; /* can be added/removed at runtime? */ -} EngramLayer; - -typedef struct EngramStore { - EngramNode* nodes; - int64_t node_count; - int64_t node_capacity; - EngramEdge* edges; - int64_t edge_count; - int64_t edge_capacity; - /* Layer registry — see engram_init_layers. The five canonical layers - * are always present; injectable layers (imprint, suit) are extended - * via engram_add_layer at runtime. layer_id values are assigned - * monotonically; removed injectable layers leave a NULL `name` slot - * (tombstone) so existing layer_id references on nodes stay stable. */ - EngramLayer* layers; - size_t layer_count; - size_t layer_capacity; -} EngramStore; - -static EngramStore* engram_global = NULL; - -/* Initialize the five canonical layers on a fresh store. Called once from - * engram_get(). Layer ids 0..4 are reserved; runtime-injected imprint/suit - * layers (engram_add_layer) get ids 5+. */ -static void engram_init_layers(EngramStore* g) { - g->layer_capacity = 16; - g->layers = calloc(g->layer_capacity, sizeof(EngramLayer)); - if (!g->layers) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - g->layer_count = 0; - - /* Layer 0 — safety. Structural floor. Non-suppressible; transparent - * (filtered out of introspection but still shapes output); not - * runtime-injectable. */ - g->layers[g->layer_count++] = (EngramLayer){ - .layer_id = ENGRAM_LAYER_SAFETY, - .name = el_strdup_persist("safety"), - .activation_priority = 0, - .suppressible = 0, - .transparent = 1, - .injectable = 0 - }; - /* Layer 1 — core-identity. The default home for legacy nodes. */ - g->layers[g->layer_count++] = (EngramLayer){ - .layer_id = ENGRAM_LAYER_CORE_IDENTITY, - .name = el_strdup_persist("core-identity"), - .activation_priority = 10, - .suppressible = 1, - .transparent = 0, - .injectable = 0 - }; - /* Layer 2 — domain-knowledge. */ - g->layers[g->layer_count++] = (EngramLayer){ - .layer_id = ENGRAM_LAYER_DOMAIN, - .name = el_strdup_persist("domain-knowledge"), - .activation_priority = 20, - .suppressible = 1, - .transparent = 0, - .injectable = 0 - }; - /* Layer 3 — imprint. Injectable: an imprint package adds/removes this - * layer (and the nodes assigned to it) as a unit. */ - g->layers[g->layer_count++] = (EngramLayer){ - .layer_id = ENGRAM_LAYER_IMPRINT, - .name = el_strdup_persist("imprint"), - .activation_priority = 30, - .suppressible = 1, - .transparent = 0, - .injectable = 1 - }; - /* Layer 4 — suit. Injectable: a Suit overlays domain skill (e.g. - * "enterprise advisor", "divorce lawyer") and can be detached. */ - g->layers[g->layer_count++] = (EngramLayer){ - .layer_id = ENGRAM_LAYER_SUIT, - .name = el_strdup_persist("suit"), - .activation_priority = 40, - .suppressible = 1, - .transparent = 0, - .injectable = 1 - }; -} - -static EngramStore* engram_get(void) { - if (engram_global) return engram_global; - engram_global = calloc(1, sizeof(EngramStore)); - if (!engram_global) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - engram_global->node_capacity = 16; - engram_global->nodes = calloc((size_t)engram_global->node_capacity, sizeof(EngramNode)); - engram_global->edge_capacity = 16; - engram_global->edges = calloc((size_t)engram_global->edge_capacity, sizeof(EngramEdge)); - engram_init_layers(engram_global); - return engram_global; -} - -/* Resolve a layer record by id. Returns NULL if no layer with that id - * exists (e.g. a removed injectable layer or a malformed snapshot). */ -static EngramLayer* engram_find_layer(uint32_t layer_id) { - EngramStore* g = engram_get(); - for (size_t i = 0; i < g->layer_count; i++) { - EngramLayer* L = &g->layers[i]; - if (!L->name) continue; /* tombstone for removed injectable layer */ - if (L->layer_id == layer_id) return L; - } - return NULL; -} - -/* Resolve a layer record by name. Returns NULL if not found. */ -static EngramLayer* engram_find_layer_by_name(const char* name) { - if (!name || !*name) return NULL; - EngramStore* g = engram_get(); - for (size_t i = 0; i < g->layer_count; i++) { - EngramLayer* L = &g->layers[i]; - if (!L->name) continue; - if (strcmp(L->name, name) == 0) return L; - } - return NULL; -} - -/* Allocate the next layer id. Skips ids that are still in use. */ -static uint32_t engram_next_layer_id(void) { - EngramStore* g = engram_get(); - uint32_t maxid = 0; - for (size_t i = 0; i < g->layer_count; i++) { - if (g->layers[i].layer_id > maxid) maxid = g->layers[i].layer_id; - } - return maxid + 1; -} - -/* Whether a node in `layer_id` may be silenced by inhibitory edges in pass 2. */ -static int engram_layer_is_suppressible(uint32_t layer_id) { - EngramLayer* L = engram_find_layer(layer_id); - if (!L) return 1; /* unknown layer → safe default: standard suppression */ - return L->suppressible ? 1 : 0; -} - -/* Whether a layer is transparent (its content shapes output but is filtered - * from introspection queries). Currently used to mark Layer 0 as invisible - * to "what do you know about yourself" lookups while still letting it - * dominate the prompt context. */ -static int engram_layer_is_transparent(uint32_t layer_id) { - EngramLayer* L = engram_find_layer(layer_id); - if (!L) return 0; - return L->transparent ? 1 : 0; -} - -static int64_t engram_now_ms(void) { - struct timeval tv; gettimeofday(&tv, NULL); - return (int64_t)tv.tv_sec * 1000LL + (int64_t)tv.tv_usec / 1000LL; -} - -static EngramNode* engram_find_node(const char* id) { - if (!id) return NULL; - EngramStore* g = engram_get(); - for (int64_t i = 0; i < g->node_count; i++) { - if (g->nodes[i].id && strcmp(g->nodes[i].id, id) == 0) return &g->nodes[i]; - } - return NULL; -} - -static int64_t engram_find_node_index(const char* id) { - if (!id) return -1; - EngramStore* g = engram_get(); - for (int64_t i = 0; i < g->node_count; i++) { - if (g->nodes[i].id && strcmp(g->nodes[i].id, id) == 0) return i; - } - return -1; -} - -static void engram_grow_nodes(void) { - EngramStore* g = engram_get(); - if (g->node_count < g->node_capacity) return; - int64_t nc = g->node_capacity * 2; - g->nodes = realloc(g->nodes, (size_t)nc * sizeof(EngramNode)); - if (!g->nodes) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - memset(g->nodes + g->node_capacity, 0, - (size_t)(nc - g->node_capacity) * sizeof(EngramNode)); - g->node_capacity = nc; -} - -static void engram_grow_edges(void) { - EngramStore* g = engram_get(); - if (g->edge_count < g->edge_capacity) return; - int64_t nc = g->edge_capacity * 2; - g->edges = realloc(g->edges, (size_t)nc * sizeof(EngramEdge)); - if (!g->edges) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - memset(g->edges + g->edge_capacity, 0, - (size_t)(nc - g->edge_capacity) * sizeof(EngramEdge)); - g->edge_capacity = nc; -} - -/* Build a fresh UUID string. Reuses uuid_new but takes the underlying char*. */ -static char* engram_new_id(void) { - el_val_t v = uuid_new(); - const char* s = EL_CSTR(v); - return el_strdup(s ? s : ""); -} - -/* Convert a node into an ElMap of its fields. */ -static el_val_t engram_node_to_map(const EngramNode* n) { - el_val_t m = el_map_new(0); - m = el_map_set(m, EL_STR(el_strdup("id")), EL_STR(el_strdup(n->id ? n->id : ""))); - m = el_map_set(m, EL_STR(el_strdup("content")), EL_STR(el_strdup(n->content ? n->content : ""))); - m = el_map_set(m, EL_STR(el_strdup("node_type")), EL_STR(el_strdup(n->node_type ? n->node_type : ""))); - m = el_map_set(m, EL_STR(el_strdup("label")), EL_STR(el_strdup(n->label ? n->label : ""))); - m = el_map_set(m, EL_STR(el_strdup("tier")), EL_STR(el_strdup(n->tier ? n->tier : "Working"))); - m = el_map_set(m, EL_STR(el_strdup("tags")), EL_STR(el_strdup(n->tags ? n->tags : ""))); - m = el_map_set(m, EL_STR(el_strdup("metadata")), EL_STR(el_strdup(n->metadata ? n->metadata : "{}"))); - m = el_map_set(m, EL_STR(el_strdup("salience")), el_from_float(n->salience)); - m = el_map_set(m, EL_STR(el_strdup("importance")), el_from_float(n->importance)); - m = el_map_set(m, EL_STR(el_strdup("confidence")), el_from_float(n->confidence)); - m = el_map_set(m, EL_STR(el_strdup("temporal_decay_rate")), el_from_float(n->temporal_decay_rate)); - m = el_map_set(m, EL_STR(el_strdup("activation_count")), (el_val_t)n->activation_count); - m = el_map_set(m, EL_STR(el_strdup("last_activated")), (el_val_t)n->last_activated); - m = el_map_set(m, EL_STR(el_strdup("created_at")), (el_val_t)n->created_at); - m = el_map_set(m, EL_STR(el_strdup("updated_at")), (el_val_t)n->updated_at); - m = el_map_set(m, EL_STR(el_strdup("background_activation")), el_from_float(n->background_activation)); - m = el_map_set(m, EL_STR(el_strdup("working_memory_weight")), el_from_float(n->working_memory_weight)); - m = el_map_set(m, EL_STR(el_strdup("suppression_count")), (el_val_t)n->suppression_count); - m = el_map_set(m, EL_STR(el_strdup("layer_id")), (el_val_t)(int64_t)n->layer_id); - return m; -} - -/* (Node JSON serialization is provided by `engram_emit_node_json` further - * down in the persistence section — reused by the *_json builtins below.) */ -static void engram_emit_node_json(JsonBuf* b, const EngramNode* n); -static void engram_emit_edge_json(JsonBuf* b, const EngramEdge* e); - -/* Salience may arrive either as a float bit-pattern or as a small integer - * (e.g. 1, meaning 1.0). Heuristic: if interpreted as double it's in - * [0.0, 100.0] use it; otherwise treat as int and convert. */ -static double engram_decode_score(el_val_t v) { - double f = el_to_float(v); - if (!isnan(f) && !isinf(f) && f >= 0.0 && f <= 100.0) return f; - int64_t n = (int64_t)v; - return (double)n; -} - -static char* engram_first_n_chars(const char* s, size_t n) { - if (!s) return el_strdup(""); - size_t l = strlen(s); - if (l > n) l = n; - char* out = el_strbuf(l); - memcpy(out, s, l); - out[l] = '\0'; - return out; -} - -el_val_t engram_node(el_val_t content, el_val_t node_type, el_val_t salience) { - EngramStore* g = engram_get(); - engram_grow_nodes(); - EngramNode* n = &g->nodes[g->node_count]; - memset(n, 0, sizeof(*n)); - n->id = engram_new_id(); - const char* c = EL_CSTR(content); - const char* nt = EL_CSTR(node_type); - n->content = el_strdup(c ? c : ""); - n->node_type = el_strdup(nt && *nt ? nt : "Memory"); - n->label = engram_first_n_chars(c, 60); - n->tier = el_strdup("Working"); - n->tags = el_strdup(""); - n->metadata = el_strdup("{}"); - n->salience = engram_decode_score(salience); - if (n->salience <= 0.0 || n->salience > 1.0) n->salience = 0.5; - n->importance = 0.5; - n->confidence = 1.0; - n->temporal_decay_rate = 0.0; /* 0 = use global default ENGRAM_DECAY_LAMBDA */ - n->activation_count = 0; - int64_t now = engram_now_ms(); - n->last_activated = now; - n->created_at = now; - n->updated_at = now; - n->layer_id = ENGRAM_LAYER_DEFAULT; - g->node_count++; - return el_wrap_str(el_strdup(n->id)); -} - -el_val_t engram_node_full(el_val_t content, el_val_t node_type, el_val_t label, - el_val_t salience, el_val_t importance, el_val_t confidence, - el_val_t tier, el_val_t tags) { - EngramStore* g = engram_get(); - engram_grow_nodes(); - EngramNode* n = &g->nodes[g->node_count]; - memset(n, 0, sizeof(*n)); - n->id = engram_new_id(); - const char* c = EL_CSTR(content); - const char* nt = EL_CSTR(node_type); - const char* lb = EL_CSTR(label); - const char* ti = EL_CSTR(tier); - const char* tg = EL_CSTR(tags); - n->content = el_strdup(c ? c : ""); - n->node_type = el_strdup(nt && *nt ? nt : "Memory"); - n->label = el_strdup(lb && *lb ? lb : (c ? engram_first_n_chars(c, 60) : "")); - n->tier = el_strdup(ti && *ti ? ti : "Working"); - n->tags = el_strdup(tg ? tg : ""); - n->metadata = el_strdup("{}"); - n->salience = engram_decode_score(salience); - n->importance = engram_decode_score(importance); - n->confidence = engram_decode_score(confidence); - if (n->salience <= 0.0 || n->salience > 1.0) n->salience = 0.5; - if (n->importance <= 0.0 || n->importance > 1.0) n->importance = 0.5; - if (n->confidence <= 0.0 || n->confidence > 1.0) n->confidence = 1.0; - n->temporal_decay_rate = 0.0; /* 0 = use global default ENGRAM_DECAY_LAMBDA */ - n->activation_count = 0; - int64_t now = engram_now_ms(); - n->last_activated = now; - n->created_at = now; - n->updated_at = now; - n->layer_id = ENGRAM_LAYER_DEFAULT; - g->node_count++; - return el_wrap_str(el_strdup(n->id)); -} - -/* engram_node_layered — like engram_node_full but with explicit layer - * assignment and an additional `status` slot reserved for callers that - * track lifecycle state in metadata. The signature mirrors the public API - * defined in the layered consciousness design doc: - * - * engram_node_layered(content, node_type, label, - * salience, certainty, confidence, - * status, tags, layer_id) - * - * `certainty` is folded into `importance` (it occupies the same axis in - * the existing schema). `status` is recorded under metadata.status; an - * empty status leaves metadata as the default "{}". - * - * If `layer_id` does not resolve to a known layer the call falls back to - * ENGRAM_LAYER_DEFAULT — better to keep the node addressable than to drop - * it because of a stale layer reference. Callers wanting strict validation - * should engram_list_layers first. */ -el_val_t engram_node_layered(el_val_t content, el_val_t node_type, el_val_t label, - el_val_t salience, el_val_t certainty, el_val_t confidence, - el_val_t status, el_val_t tags, el_val_t layer_id) { - EngramStore* g = engram_get(); - engram_grow_nodes(); - EngramNode* n = &g->nodes[g->node_count]; - memset(n, 0, sizeof(*n)); - n->id = engram_new_id(); - const char* c = EL_CSTR(content); - const char* nt = EL_CSTR(node_type); - const char* lb = EL_CSTR(label); - const char* tg = EL_CSTR(tags); - const char* st = EL_CSTR(status); - n->content = el_strdup(c ? c : ""); - n->node_type = el_strdup(nt && *nt ? nt : "Memory"); - n->label = el_strdup(lb && *lb ? lb : (c ? engram_first_n_chars(c, 60) : "")); - n->tier = el_strdup("Working"); - n->tags = el_strdup(tg ? tg : ""); - if (st && *st) { - /* Minimal metadata payload: {"status":"..."}. Keep it cheap so - * callers using `status` don't pay JSON parse cost on every read. */ - size_t sl = strlen(st) + 16; - char* meta = el_strbuf(sl); - snprintf(meta, sl, "{\"status\":\"%s\"}", st); - n->metadata = meta; - } else { - n->metadata = el_strdup("{}"); - } - n->salience = engram_decode_score(salience); - n->importance = engram_decode_score(certainty); - n->confidence = engram_decode_score(confidence); - if (n->salience <= 0.0 || n->salience > 1.0) n->salience = 0.5; - if (n->importance <= 0.0 || n->importance > 1.0) n->importance = 0.5; - if (n->confidence <= 0.0 || n->confidence > 1.0) n->confidence = 1.0; - n->temporal_decay_rate = 0.0; - n->activation_count = 0; - int64_t now = engram_now_ms(); - n->last_activated = now; - n->created_at = now; - n->updated_at = now; - /* Resolve layer assignment. Caller passes either a numeric layer_id or - * a stringified id; el_to_float / int cast tolerates both. */ - int64_t lid = (int64_t)layer_id; - if (lid < 0) lid = (int64_t)ENGRAM_LAYER_DEFAULT; - if (!engram_find_layer((uint32_t)lid)) lid = (int64_t)ENGRAM_LAYER_DEFAULT; - n->layer_id = (uint32_t)lid; - g->node_count++; - return el_wrap_str(el_strdup(n->id)); -} - -/* ── Layer registry public API ────────────────────────────────────────────── - * - * The five canonical layers are seeded at engram_get() initialization. - * Runtime code (typically imprint/suit injection logic at the EL level) - * can extend the registry with engram_add_layer() — only layers marked - * `injectable=1` may be removed via engram_remove_layer(). Removing a - * layer leaves a tombstone slot so existing layer_id references on nodes - * stay valid; orphaned references resolve to "unknown layer" and inherit - * the default suppression behavior. - */ - -/* engram_add_layer — register a new layer at runtime. - * Returns the assigned layer_id as an el_val_t int (cast back via int64_t). - * Conflicting names are rejected (returns 0). */ -el_val_t engram_add_layer(el_val_t name, el_val_t priority, el_val_t suppressible, - el_val_t transparent, el_val_t injectable) { - EngramStore* g = engram_get(); - const char* nm = EL_CSTR(name); - if (!nm || !*nm) return (el_val_t)0; - if (engram_find_layer_by_name(nm)) { - /* Name collision — return existing id so callers are idempotent. */ - return (el_val_t)(int64_t)engram_find_layer_by_name(nm)->layer_id; - } - if (g->layer_count >= g->layer_capacity) { - size_t nc = g->layer_capacity ? g->layer_capacity * 2 : 16; - EngramLayer* grown = realloc(g->layers, nc * sizeof(EngramLayer)); - if (!grown) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - memset(grown + g->layer_capacity, 0, - (nc - g->layer_capacity) * sizeof(EngramLayer)); - g->layers = grown; - g->layer_capacity = nc; - } - EngramLayer* L = &g->layers[g->layer_count++]; - L->layer_id = engram_next_layer_id(); - L->name = el_strdup_persist(nm); - L->activation_priority = (uint32_t)(int64_t)priority; - L->suppressible = (int)(int64_t)suppressible ? 1 : 0; - L->transparent = (int)(int64_t)transparent ? 1 : 0; - L->injectable = (int)(int64_t)injectable ? 1 : 0; - return (el_val_t)(int64_t)L->layer_id; -} - -/* engram_remove_layer — remove an injectable layer by id. - * Built-in (non-injectable) layers cannot be removed. Nodes still tagged - * with the removed layer's id keep their tag but resolve to "unknown - * layer" thereafter and inherit standard (suppressible) behavior. - * Returns 1 on success, 0 on failure (unknown id, non-injectable). */ -el_val_t engram_remove_layer(el_val_t layer_id) { - EngramStore* g = engram_get(); - int64_t lid = (int64_t)layer_id; - for (size_t i = 0; i < g->layer_count; i++) { - EngramLayer* L = &g->layers[i]; - if (!L->name) continue; - if ((int64_t)L->layer_id != lid) continue; - if (!L->injectable) return (el_val_t)0; - free(L->name); - L->name = NULL; /* tombstone */ - /* Leave layer_id, priority, flags intact so debug snapshots can - * still distinguish "removed at runtime" from "never existed". */ - return (el_val_t)1; - } - return (el_val_t)0; -} - -/* engram_list_layers — enumerate the active layer registry. - * Returns an ElList of maps, one per non-tombstone layer, sorted by - * activation_priority ascending (deepest layer first). */ -el_val_t engram_list_layers(void) { - EngramStore* g = engram_get(); - el_val_t lst = el_list_empty(); - if (g->layer_count == 0) return lst; - /* Build an index sorted by activation_priority ascending. */ - size_t* idx = malloc(g->layer_count * sizeof(size_t)); - if (!idx) return lst; - size_t live = 0; - for (size_t i = 0; i < g->layer_count; i++) { - if (g->layers[i].name) idx[live++] = i; - } - /* Insertion sort — N is small (≤ a few dozen layers). */ - for (size_t i = 1; i < live; i++) { - size_t key = idx[i]; - uint32_t kp = g->layers[key].activation_priority; - size_t j = i; - while (j > 0 && g->layers[idx[j - 1]].activation_priority > kp) { - idx[j] = idx[j - 1]; - j--; - } - idx[j] = key; - } - for (size_t i = 0; i < live; i++) { - EngramLayer* L = &g->layers[idx[i]]; - el_val_t m = el_map_new(0); - m = el_map_set(m, EL_STR(el_strdup("layer_id")), - (el_val_t)(int64_t)L->layer_id); - m = el_map_set(m, EL_STR(el_strdup("name")), - EL_STR(el_strdup(L->name ? L->name : ""))); - m = el_map_set(m, EL_STR(el_strdup("activation_priority")), - (el_val_t)(int64_t)L->activation_priority); - m = el_map_set(m, EL_STR(el_strdup("suppressible")), - (el_val_t)(int64_t)(L->suppressible ? 1 : 0)); - m = el_map_set(m, EL_STR(el_strdup("transparent")), - (el_val_t)(int64_t)(L->transparent ? 1 : 0)); - m = el_map_set(m, EL_STR(el_strdup("injectable")), - (el_val_t)(int64_t)(L->injectable ? 1 : 0)); - lst = el_list_append(lst, m); - } - free(idx); - return lst; -} - -el_val_t engram_get_node(el_val_t id) { - const char* sid = EL_CSTR(id); - EngramNode* n = engram_find_node(sid); - if (!n) return el_map_new(0); - return engram_node_to_map(n); -} - -void engram_strengthen(el_val_t node_id) { - const char* sid = EL_CSTR(node_id); - EngramNode* n = engram_find_node(sid); - if (!n) return; - n->salience += 0.05; - if (n->salience > 1.0) n->salience = 1.0; - n->activation_count++; - n->last_activated = engram_now_ms(); - n->updated_at = n->last_activated; -} - -void engram_forget(el_val_t node_id) { - const char* sid = EL_CSTR(node_id); - if (!sid) return; - EngramStore* g = engram_get(); - int64_t idx = engram_find_node_index(sid); - if (idx < 0) return; - /* Free node strings */ - EngramNode* n = &g->nodes[idx]; - free(n->id); free(n->content); free(n->node_type); free(n->label); - free(n->tier); free(n->tags); free(n->metadata); - /* Shift remaining nodes down */ - for (int64_t i = idx + 1; i < g->node_count; i++) { - g->nodes[i - 1] = g->nodes[i]; - } - g->node_count--; - memset(&g->nodes[g->node_count], 0, sizeof(EngramNode)); - /* Remove all incident edges */ - int64_t w = 0; - for (int64_t r = 0; r < g->edge_count; r++) { - EngramEdge* e = &g->edges[r]; - int incident = (e->from_id && strcmp(e->from_id, sid) == 0) || - (e->to_id && strcmp(e->to_id, sid) == 0); - if (incident) { - free(e->id); free(e->from_id); free(e->to_id); - free(e->relation); free(e->metadata); - } else { - if (w != r) g->edges[w] = g->edges[r]; - w++; - } - } - g->edge_count = w; -} - -el_val_t engram_node_count(void) { - return (el_val_t)engram_get()->node_count; -} - -static int istr_contains(const char* hay, const char* needle) { - if (!hay || !needle || !*needle) return 0; - size_t nl = strlen(needle); - for (const char* p = hay; *p; p++) { - if (strncasecmp(p, needle, nl) == 0) return 1; - } - return 0; -} - -el_val_t engram_search(el_val_t query, el_val_t limit) { - EngramStore* g = engram_get(); - const char* q = EL_CSTR(query); - int64_t lim = (int64_t)limit; - if (lim <= 0) lim = 100; - el_val_t lst = el_list_empty(); - if (!q || !*q) return lst; - int64_t found = 0; - for (int64_t i = 0; i < g->node_count && found < lim; i++) { - EngramNode* n = &g->nodes[i]; - /* Filter transparent layers: nodes whose layer is `transparent=1` - * shape output but are invisible to introspection ("what do you - * know about yourself"). They still surface via engram_activate - * + engram_compile_layered_json — that's the legitimate path. */ - if (engram_layer_is_transparent(n->layer_id)) continue; - if (istr_contains(n->content, q) || - istr_contains(n->label, q) || - istr_contains(n->tags, q)) { - lst = el_list_append(lst, engram_node_to_map(n)); - found++; - } - } - return lst; -} - -/* Sort node indices by salience desc (small N, insertion sort is fine). */ -static void engram_sort_indices_by_salience(int64_t* arr, int64_t n, - const EngramNode* nodes) { - for (int64_t i = 1; i < n; i++) { - int64_t key = arr[i]; - double ks = nodes[key].salience; - int64_t j = i - 1; - while (j >= 0 && nodes[arr[j]].salience < ks) { - arr[j + 1] = arr[j]; - j--; - } - arr[j + 1] = key; - } -} - -el_val_t engram_scan_nodes(el_val_t limit, el_val_t offset) { - EngramStore* g = engram_get(); - int64_t lim = (int64_t)limit; if (lim <= 0) lim = 100; - int64_t off = (int64_t)offset; if (off < 0) off = 0; - el_val_t lst = el_list_empty(); - if (g->node_count == 0) return lst; - int64_t* idx = malloc((size_t)g->node_count * sizeof(int64_t)); - if (!idx) return lst; - /* Skip transparent layers — same introspection-filter rationale as - * engram_search above. */ - int64_t live = 0; - for (int64_t i = 0; i < g->node_count; i++) { - if (engram_layer_is_transparent(g->nodes[i].layer_id)) continue; - idx[live++] = i; - } - engram_sort_indices_by_salience(idx, live, g->nodes); - int64_t end = off + lim; - if (end > live) end = live; - for (int64_t i = off; i < end; i++) { - lst = el_list_append(lst, engram_node_to_map(&g->nodes[idx[i]])); - } - free(idx); - return lst; -} - -void engram_connect(el_val_t from_id, el_val_t to_id, el_val_t weight, el_val_t relation) { - EngramStore* g = engram_get(); - const char* f = EL_CSTR(from_id); - const char* t = EL_CSTR(to_id); - const char* r = EL_CSTR(relation); - if (!f || !t) return; - engram_grow_edges(); - EngramEdge* e = &g->edges[g->edge_count]; - memset(e, 0, sizeof(*e)); - e->id = engram_new_id(); - e->from_id = el_strdup(f); - e->to_id = el_strdup(t); - e->relation = el_strdup(r && *r ? r : "associate"); - e->metadata = el_strdup("{}"); - e->weight = engram_decode_score(weight); - if (e->weight <= 0.0 || e->weight > 1.0) e->weight = 0.5; - e->confidence = 1.0; - int64_t now = engram_now_ms(); - e->created_at = now; - e->updated_at = now; - e->last_fired = 0; - e->layer_id = ENGRAM_LAYER_DEFAULT; - g->edge_count++; -} - -el_val_t engram_edge_between(el_val_t from_id, el_val_t to_id) { - EngramStore* g = engram_get(); - const char* f = EL_CSTR(from_id); - const char* t = EL_CSTR(to_id); - if (!f || !t) return 0; - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - if (e->from_id && e->to_id && - strcmp(e->from_id, f) == 0 && strcmp(e->to_id, t) == 0) return 1; - } - return 0; -} - -/* Reserved helper: edge -> ElMap. Kept around for future builtins. */ -static el_val_t engram_edge_to_map(const EngramEdge* e) __attribute__((unused)); -static el_val_t engram_edge_to_map(const EngramEdge* e) { - el_val_t m = el_map_new(0); - m = el_map_set(m, EL_STR(el_strdup("id")), EL_STR(el_strdup(e->id ? e->id : ""))); - m = el_map_set(m, EL_STR(el_strdup("from_id")), EL_STR(el_strdup(e->from_id ? e->from_id : ""))); - m = el_map_set(m, EL_STR(el_strdup("to_id")), EL_STR(el_strdup(e->to_id ? e->to_id : ""))); - m = el_map_set(m, EL_STR(el_strdup("relation")), EL_STR(el_strdup(e->relation ? e->relation : ""))); - m = el_map_set(m, EL_STR(el_strdup("metadata")), EL_STR(el_strdup(e->metadata ? e->metadata : "{}"))); - m = el_map_set(m, EL_STR(el_strdup("weight")), el_from_float(e->weight)); - m = el_map_set(m, EL_STR(el_strdup("confidence")), el_from_float(e->confidence)); - m = el_map_set(m, EL_STR(el_strdup("created_at")), (el_val_t)e->created_at); - m = el_map_set(m, EL_STR(el_strdup("updated_at")), (el_val_t)e->updated_at); - m = el_map_set(m, EL_STR(el_strdup("last_fired")), (el_val_t)e->last_fired); - m = el_map_set(m, EL_STR(el_strdup("inhibitory")), (el_val_t)(e->inhibitory ? 1 : 0)); - m = el_map_set(m, EL_STR(el_strdup("layer_id")), (el_val_t)(int64_t)e->layer_id); - return m; -} - -el_val_t engram_neighbors(el_val_t node_id) { - EngramStore* g = engram_get(); - const char* sid = EL_CSTR(node_id); - el_val_t lst = el_list_empty(); - if (!sid) return lst; - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - const char* other = NULL; - if (e->from_id && strcmp(e->from_id, sid) == 0) other = e->to_id; - else if (e->to_id && strcmp(e->to_id, sid) == 0) other = e->from_id; - if (!other) continue; - EngramNode* n = engram_find_node(other); - if (n) lst = el_list_append(lst, engram_node_to_map(n)); - } - return lst; -} - -el_val_t engram_neighbors_filtered(el_val_t node_id, el_val_t max_depth, el_val_t direction) { - EngramStore* g = engram_get(); - const char* sid = EL_CSTR(node_id); - int64_t md = (int64_t)max_depth; if (md <= 0) md = 1; - const char* dir = EL_CSTR(direction); /* "out" | "in" | "both" (default) */ - el_val_t lst = el_list_empty(); - if (!sid || g->node_count == 0) return lst; - int64_t start = engram_find_node_index(sid); - if (start < 0) return lst; - /* BFS with depth tracking */ - int64_t* visited = calloc((size_t)g->node_count, sizeof(int64_t)); - int64_t* queue = calloc((size_t)g->node_count, sizeof(int64_t)); - int64_t* depths = calloc((size_t)g->node_count, sizeof(int64_t)); - if (!visited || !queue || !depths) { - free(visited); free(queue); free(depths); return lst; - } - int64_t qh = 0, qt = 0; - queue[qt++] = start; - visited[start] = 1; - depths[start] = 0; - while (qh < qt) { - int64_t cur = queue[qh++]; - const char* cur_id = g->nodes[cur].id; - int64_t cur_depth = depths[cur]; - if (cur_depth >= md) continue; - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - const char* other = NULL; - int outgoing = e->from_id && strcmp(e->from_id, cur_id) == 0; - int incoming = e->to_id && strcmp(e->to_id, cur_id) == 0; - if (dir && strcmp(dir, "out") == 0 && !outgoing) continue; - if (dir && strcmp(dir, "in") == 0 && !incoming) continue; - if (outgoing) other = e->to_id; - else if (incoming) other = e->from_id; - else continue; - int64_t oi = engram_find_node_index(other); - if (oi < 0 || visited[oi]) continue; - visited[oi] = 1; - depths[oi] = cur_depth + 1; - queue[qt++] = oi; - } - } - /* Emit all visited except the seed */ - for (int64_t i = 0; i < g->node_count; i++) { - if (visited[i] && i != start) { - lst = el_list_append(lst, engram_node_to_map(&g->nodes[i])); - } - } - free(visited); free(queue); free(depths); - return lst; -} - -el_val_t engram_edge_count(void) { - return (el_val_t)engram_get()->edge_count; -} - -/* Compute temporal decay factor for a node given current time. - * effective contribution = salience * exp(-lambda * age_hours / T_half) - * Clamped to [0.05, 1.0] so very old nodes retain a meaningful floor. */ -static double engram_temporal_decay(const EngramNode* n, int64_t now_ms) { - int64_t age_ms = now_ms - n->last_activated; - if (age_ms <= 0) return 1.0; - double lambda = (n->temporal_decay_rate > 0.0) ? n->temporal_decay_rate - : ENGRAM_DECAY_LAMBDA; - double age_hours = (double)age_ms / 3600000.0; - double factor = exp(-lambda * age_hours / ENGRAM_T_HALF_HOURS); - if (factor < 0.05) factor = 0.05; - return factor; -} - -/* Activation dampening: high activation_count nodes are "well-known" context - * and get less marginal boost per firing. - * count=0 → 1.0, count=2 → ~0.74, count=9 → ~0.59, count=99 → ~0.43 */ -static double engram_activation_dampen(const EngramNode* n) { - return 1.0 / (1.0 + log(1.0 + (double)n->activation_count)); -} - -/* Temporal proximity bonus: boost propagation along edges connecting - * co-temporal nodes. Returns a multiplier bonus in [0, 0.2]. */ -static double engram_temporal_proximity_bonus(int64_t node_created, - int64_t seed_epoch) { - int64_t diff = node_created - seed_epoch; - if (diff < 0) diff = -diff; - if (diff < 86400000LL) return 0.20; /* within 1 day */ - if (diff < 604800000LL) return 0.10; /* within 7 days */ - return 0.0; -} - -/* ── Two-layer activation (biologically-motivated) ─────────────────────────── - * - * Layer 1 — Broad fan-out (background activation): - * BFS + spreading activation fires on ALL nodes reachable from seeds, - * regardless of relevance to the current goal. Every reachable node gets - * a background_activation score. Nothing is filtered here. Models the - * brain's massive parallel sub-threshold activation of all associated - * content in response to a stimulus. Temporal decay and activation - * dampening are applied at this layer (as before), but no threshold gate. - * - * Layer 2 — Executive filter (working memory promotion): - * A second pass asks: given the query (goal intent), attentional bias, - * and inhibitory edge topology — which background-activated nodes should - * break through into working memory? - * - * wm_weight = bg_activation * goal_bias(node, query) * confidence - * * inhibitory_suppression_factor - * - * Only nodes where wm_weight >= ENGRAM_WM_THRESHOLD are promoted to - * working memory (working_memory_weight > 0). Background-activated nodes - * that don't cross the threshold accumulate suppression_count. After - * ENGRAM_SUPPRESSION_BREAKTHROUGH consecutive suppressed turns, the node - * force-breaks through at ENGRAM_BREAKTHROUGH_WEIGHT (latent tension - * surfacing — models intrusive memory / unresolved cognitive load). - * - * Inhibitory edges: - * An edge with inhibitory=1 suppresses the TARGET node's working memory - * promotion when the SOURCE is background-activated. Background activation - * of the target is NOT affected — the node fires in layer 1. Only the - * executive filter (layer 2) is gated. Models attentional inhibition: - * "focused on code work" suppresses personal memories from surfacing - * even if they have high background_activation. - * - * Goal bias: - * A lightweight heuristic rates how well each background-activated node - * aligns with the apparent intent of the current query. Technical queries - * boost Belief/Canonical/Lesson nodes; relational queries boost Memory/ - * Entity nodes. Direct lexical overlap gives a 50% bonus. - * - * Working memory persistence (turn continuity): - * Nodes promoted in the previous turn retain a decayed working_memory_weight - * (weight *= ENGRAM_WM_DECAY) without needing re-activation. This models - * conversational thread continuity — once a topic is in working memory, - * it persists slightly into the next turn. - * - * Returns ElList of {node, activation_strength, working_memory_weight, - * epistemic_confidence, hops, promoted}. - * "promoted" = 1 if working_memory_weight > 0, 0 if background-only. - * Context compilation uses ONLY nodes with promoted=1. - * - * Temporal decay (preserved from prior implementation): - * effective_salience = salience * exp(-lambda * age_hours / T_half) - * where T_half = 168 h (one week), lambda = ln(2) - * - * Activation dampening (preserved): - * dampen = 1 / (1 + log(1 + activation_count)) - * - * Temporal proximity bonus (preserved): - * edge_strength *= (1 + tbonus) where tbonus ∈ {0, 0.10, 0.20} - * - * Per-type threshold gates apply only to working memory promotion (layer 2): - * Safety/DharmaSelf: 0.05 Canonical: 0.15 Lesson: 0.25 - * Belief/Entity: 0.30 Note/Memory/Working: 0.40 - */ - -/* Compute goal-state bias multiplier for a node given the query. - * Returns a value in [0.3, 2.0]. This is a lightweight heuristic — - * a production implementation may use LLM-derived intent classification. */ -static double engram_goal_bias(const EngramNode* n, const char* query) { - if (!query || !*query) return 1.0; - double bias = 1.0; - /* Direct lexical overlap: node content/label/tags share text with query. */ - if (istr_contains(n->content, query) || istr_contains(n->label, query) || - istr_contains(n->tags, query)) { - bias += 0.5; - } - /* Node-type resonance with query intent. */ - int technical_query = istr_contains(query, "code") || - istr_contains(query, "function") || - istr_contains(query, "implement") || - istr_contains(query, "error") || - istr_contains(query, "bug") || - istr_contains(query, "build") || - istr_contains(query, "system") || - istr_contains(query, "design") || - istr_contains(query, "architecture"); - int personal_query = istr_contains(query, "feel") || - istr_contains(query, "emotion") || - istr_contains(query, "remember") || - istr_contains(query, "personal") || - istr_contains(query, "story") || - istr_contains(query, "relationship"); - if (n->node_type) { - int is_knowledge = (strcmp(n->node_type, "Belief") == 0) || - (strcmp(n->node_type, "DharmaSelf") == 0) || - (strcmp(n->node_type, "Safety") == 0); - int is_personal = (strcmp(n->node_type, "Memory") == 0) || - (strcmp(n->node_type, "Entity") == 0); - if (technical_query && is_knowledge) bias += 0.3; - if (technical_query && is_personal) bias -= 0.3; - if (personal_query && is_personal) bias += 0.3; - if (personal_query && is_knowledge) bias -= 0.1; - } - /* Tier-based bonus: promote higher-confidence knowledge nodes. */ - if (n->tier) { - if (strcmp(n->tier, "Canonical") == 0) bias += 0.2; - if (strcmp(n->tier, "Lesson") == 0) bias += 0.1; - } - if (bias < 0.3) bias = 0.3; - if (bias > 2.0) bias = 2.0; - return bias; -} - -el_val_t engram_activate(el_val_t query, el_val_t depth) { - EngramStore* g = engram_get(); - const char* q = EL_CSTR(query); - int64_t max_depth = (int64_t)depth; if (max_depth <= 0) max_depth = 2; - el_val_t out = el_list_empty(); - if (!q || g->node_count == 0) return out; - - int64_t now_ms = engram_now_ms(); - - /* Per-node layer-1 tracking. */ - double* best_bg = calloc((size_t)g->node_count, sizeof(double)); - int64_t* best_hops = calloc((size_t)g->node_count, sizeof(int64_t)); - int* reached = calloc((size_t)g->node_count, sizeof(int)); - if (!best_bg || !best_hops || !reached) { - free(best_bg); free(best_hops); free(reached); return out; - } - - /* ── LAYER 1: broad fan-out (background activation) ───────────────── - * Find seeds, apply temporal decay + dampening, BFS with edge weights. - * Inhibitory edges propagate activation normally at this layer — they - * only gate working memory promotion in layer 2. */ - typedef struct { int64_t idx; double act; int64_t created_at; } SeedEntry; - SeedEntry* seeds = malloc((size_t)g->node_count * sizeof(SeedEntry)); - int64_t seed_count = 0; - if (!seeds) { - free(best_bg); free(best_hops); free(reached); return out; - } - for (int64_t i = 0; i < g->node_count; i++) { - EngramNode* n = &g->nodes[i]; - if (istr_contains(n->content, q) || - istr_contains(n->label, q) || - istr_contains(n->tags, q)) { - double tdecay = engram_temporal_decay(n, now_ms); - double dampen = engram_activation_dampen(n); - double act = n->salience * tdecay * dampen; - seeds[seed_count].idx = i; - seeds[seed_count].act = act; - seeds[seed_count].created_at = n->created_at; - seed_count++; - best_bg[i] = act; - best_hops[i] = 0; - reached[i] = 1; - } - } - /* Compute mean seed created_at for temporal proximity bonus. */ - int64_t seed_epoch = 0; - if (seed_count > 0) { - seed_epoch = seeds[0].created_at; - for (int64_t s = 1; s < seed_count; s++) - seed_epoch = (seed_epoch + seeds[s].created_at) / 2; - } - typedef struct { int64_t idx; int64_t hops; double act; } Frontier; - Frontier* fr = malloc((size_t)(g->node_count * (max_depth + 1)) * sizeof(Frontier) + 16 * sizeof(Frontier)); - if (!fr) { - free(best_bg); free(best_hops); free(reached); free(seeds); return out; - } - int64_t fhead = 0, ftail = 0; - int64_t fcap = (int64_t)((size_t)(g->node_count * (max_depth + 1)) + 16); - for (int64_t s = 0; s < seed_count; s++) { - if (ftail >= fcap) break; - fr[ftail].idx = seeds[s].idx; - fr[ftail].hops = 0; - fr[ftail].act = seeds[s].act; - ftail++; - } - const double SPREAD_DECAY = 0.7; - while (fhead < ftail) { - Frontier f = fr[fhead++]; - if (f.hops >= max_depth) continue; - const char* cur_id = g->nodes[f.idx].id; - for (int64_t ei = 0; ei < g->edge_count; ei++) { - EngramEdge* e = &g->edges[ei]; - const char* other = NULL; - if (e->from_id && strcmp(e->from_id, cur_id) == 0) other = e->to_id; - else if (e->to_id && strcmp(e->to_id, cur_id) == 0) other = e->from_id; - else continue; - int64_t oi = engram_find_node_index(other); - if (oi < 0) continue; - EngramNode* on = &g->nodes[oi]; - double tbonus = engram_temporal_proximity_bonus(on->created_at, seed_epoch); - double tdecay = engram_temporal_decay(on, now_ms); - double dampen = engram_activation_dampen(on); - double new_act = f.act * e->weight * SPREAD_DECAY * (1.0 + tbonus) - * tdecay * dampen; - int64_t new_hops = f.hops + 1; - if (!reached[oi] || new_act > best_bg[oi]) { - best_bg[oi] = new_act; - best_hops[oi] = new_hops; - reached[oi] = 1; - if (ftail < fcap) { - fr[ftail].idx = oi; - fr[ftail].hops = new_hops; - fr[ftail].act = new_act; - ftail++; - } - } - } - } - /* Persist layer-1 background_activation to node store. */ - for (int64_t i = 0; i < g->node_count; i++) { - g->nodes[i].background_activation = reached[i] ? best_bg[i] : 0.0; - } - - /* ── PASS 2: executive filter → working memory promotion ──────────── */ - /* Step A: collect inhibitory suppressions from fired inhibitory edges. - * Layered consciousness: inhibition is ONLY recorded against targets - * whose layer is `suppressible == 1`. Nodes in non-suppressible layers - * (Layer 0 / safety) ignore inhibitory edges entirely — their working - * memory weight cannot be silenced by attentional suppression. */ - double* inhibition = calloc((size_t)g->node_count, sizeof(double)); - if (!inhibition) { - free(best_bg); free(best_hops); free(reached); free(seeds); free(fr); - return out; - } - for (int64_t ei = 0; ei < g->edge_count; ei++) { - EngramEdge* e = &g->edges[ei]; - if (!e->inhibitory) continue; - int64_t src = engram_find_node_index(e->from_id); - int64_t tgt = engram_find_node_index(e->to_id); - if (src < 0 || tgt < 0) continue; - if (!reached[src] || best_bg[src] <= 0.0) continue; - /* Skip if target layer is non-suppressible: Layer 0 / safety nodes - * are immune to inhibitory edges from any source. The pass-3 - * override below also force-promotes them, but recording inhibition - * against them at all would be wasted work and could confuse - * downstream debugging output. */ - if (!engram_layer_is_suppressible(g->nodes[tgt].layer_id)) continue; - /* Inhibition strength proportional to source background activation - * and edge weight. Takes the maximum if multiple inhibitory edges - * target the same node. */ - double inh = best_bg[src] * e->weight; - if (inh > inhibition[tgt]) inhibition[tgt] = inh; - } - /* Step B: compute working_memory_weight per candidate node. */ - double* wm_weights = calloc((size_t)g->node_count, sizeof(double)); - if (!wm_weights) { - free(best_bg); free(best_hops); free(reached); free(seeds); - free(fr); free(inhibition); return out; - } - for (int64_t i = 0; i < g->node_count; i++) { - if (!reached[i] || best_bg[i] <= 0.0) continue; - EngramNode* n = &g->nodes[i]; - /* Per-type threshold: safety nodes break through more easily. */ - double type_threshold = engram_type_threshold(n->node_type, n->tier); - /* Goal bias weights the node's relevance to current intent. */ - double bias = engram_goal_bias(n, q); - /* Raw working memory score. */ - double raw_wm = best_bg[i] * bias * n->confidence; - /* Apply inhibitory suppression. Full inhibition → scale by factor. */ - double inh = inhibition[i]; - if (inh > 1.0) inh = 1.0; - double suppress = 1.0 - (1.0 - ENGRAM_INHIBITION_FACTOR) * inh; - raw_wm *= suppress; - /* Threshold gate: must exceed per-type threshold to enter working - * memory. Type threshold replaces the old flat 0.2 filter. */ - if (raw_wm >= type_threshold) { - wm_weights[i] = raw_wm > 1.0 ? 1.0 : raw_wm; - if (n->suppression_count > 0) n->suppression_count = 0; - } else { - /* Node didn't make it through — increment suppression counter. - * After N consecutive suppressions: force breakthrough. */ - n->suppression_count++; - if (n->suppression_count >= ENGRAM_SUPPRESSION_BREAKTHROUGH) { - wm_weights[i] = ENGRAM_BREAKTHROUGH_WEIGHT; - n->suppression_count = 0; - } else { - wm_weights[i] = 0.0; - } - } - } - /* ── PASS 3: Layer 0 override (the sacred fire) ───────────────────── - * Every node in a non-suppressible layer that received any background - * activation is force-promoted to AT LEAST ENGRAM_LAYER0_OVERRIDE_WEIGHT. - * This runs LAST and overrides whatever Pass 2 decided — Layer 0 cannot - * be silenced by inhibitory edges, by goal-bias misalignment, by - * confidence weighting, or by per-type threshold gates. If the seed - * fan-out reached a structural-floor node, that node surfaces. - * - * Note: this also clears the suppression_count when an override fires, - * since the node DID surface this turn — it just took the override path - * rather than the standard threshold path. Without this, a Layer 0 - * node with persistent inhibitory pressure would accumulate - * suppression_count forever and never reach the breakthrough state. */ - for (int64_t i = 0; i < g->node_count; i++) { - if (!reached[i] || best_bg[i] <= 0.0) continue; - EngramNode* n = &g->nodes[i]; - if (engram_layer_is_suppressible(n->layer_id)) continue; - if (wm_weights[i] < ENGRAM_LAYER0_OVERRIDE_WEIGHT) { - wm_weights[i] = ENGRAM_LAYER0_OVERRIDE_WEIGHT; - } - n->suppression_count = 0; - } - - /* Persist working_memory_weight (post Pass 3) to node store. */ - for (int64_t i = 0; i < g->node_count; i++) { - g->nodes[i].working_memory_weight = wm_weights[i]; - } - - /* ── Collect all background-activated nodes for the return value ──── - * Callers see both layers. Context compilation uses only promoted nodes - * (working_memory_weight > 0). Sort: promoted first by wm_weight desc, - * then background-only by background_activation desc. */ - typedef struct { int64_t idx; double bg; double wm; double epist; int64_t hops; } Result; - Result* results = malloc((size_t)g->node_count * sizeof(Result)); - int64_t rcount = 0; - if (!results) { - free(best_bg); free(best_hops); free(reached); free(seeds); - free(fr); free(inhibition); free(wm_weights); return out; - } - for (int64_t i = 0; i < g->node_count; i++) { - if (!reached[i]) continue; - double epist = best_bg[i] * g->nodes[i].confidence; - /* Include if promoted to working memory OR if background activation - * is meaningful enough to report (epist >= 0.1). */ - if (epist < 0.1 && wm_weights[i] <= 0.0) continue; - results[rcount].idx = i; - results[rcount].bg = best_bg[i]; - results[rcount].wm = wm_weights[i]; - results[rcount].epist = epist; - results[rcount].hops = best_hops[i]; - rcount++; - } - /* Sort: promoted nodes first (by wm_weight desc), then background-only - * by background_activation desc. */ - for (int64_t i = 1; i < rcount; i++) { - Result key = results[i]; - int64_t j = i - 1; - while (j >= 0 && (results[j].wm < key.wm || - (results[j].wm == key.wm && results[j].bg < key.bg))) { - results[j + 1] = results[j]; - j--; - } - results[j + 1] = key; - } - for (int64_t i = 0; i < rcount; i++) { - el_val_t entry = el_map_new(0); - entry = el_map_set(entry, EL_STR(el_strdup("node")), - engram_node_to_map(&g->nodes[results[i].idx])); - entry = el_map_set(entry, EL_STR(el_strdup("activation_strength")), - el_from_float(results[i].bg)); - entry = el_map_set(entry, EL_STR(el_strdup("working_memory_weight")), - el_from_float(results[i].wm)); - entry = el_map_set(entry, EL_STR(el_strdup("epistemic_confidence")), - el_from_float(results[i].epist)); - entry = el_map_set(entry, EL_STR(el_strdup("hops")), - (el_val_t)results[i].hops); - entry = el_map_set(entry, EL_STR(el_strdup("promoted")), - (el_val_t)(results[i].wm > 0.0 ? 1 : 0)); - out = el_list_append(out, entry); - } - free(best_bg); free(best_hops); free(reached); - free(seeds); free(fr); free(inhibition); free(wm_weights); free(results); - return out; -} - -/* ── Engram persistence (JSON snapshot) ─────────────────────────────────── */ - -static void engram_emit_node_json(JsonBuf* b, const EngramNode* n) { - jb_putc(b, '{'); - jb_puts(b, "\"id\":"); jb_emit_escaped(b, n->id ? n->id : ""); - jb_puts(b, ",\"content\":"); jb_emit_escaped(b, n->content ? n->content : ""); - jb_puts(b, ",\"node_type\":"); jb_emit_escaped(b, n->node_type ? n->node_type : ""); - jb_puts(b, ",\"label\":"); jb_emit_escaped(b, n->label ? n->label : ""); - jb_puts(b, ",\"tier\":"); jb_emit_escaped(b, n->tier ? n->tier : "Working"); - jb_puts(b, ",\"tags\":"); jb_emit_escaped(b, n->tags ? n->tags : ""); - jb_puts(b, ",\"metadata\":"); jb_emit_escaped(b, n->metadata ? n->metadata : "{}"); - char tmp[80]; - snprintf(tmp, sizeof(tmp), ",\"salience\":%g", n->salience); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"importance\":%g", n->importance); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"confidence\":%g", n->confidence); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"temporal_decay_rate\":%g", n->temporal_decay_rate); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"activation_count\":%lld", (long long)n->activation_count); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"last_activated\":%lld", (long long)n->last_activated); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"created_at\":%lld", (long long)n->created_at); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"updated_at\":%lld", (long long)n->updated_at); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"background_activation\":%g", n->background_activation); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"working_memory_weight\":%g", n->working_memory_weight); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"suppression_count\":%d", n->suppression_count); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"layer_id\":%u", n->layer_id); jb_puts(b, tmp); - jb_putc(b, '}'); -} - -static void engram_emit_edge_json(JsonBuf* b, const EngramEdge* e) { - jb_putc(b, '{'); - jb_puts(b, "\"id\":"); jb_emit_escaped(b, e->id ? e->id : ""); - jb_puts(b, ",\"from_id\":"); jb_emit_escaped(b, e->from_id ? e->from_id : ""); - jb_puts(b, ",\"to_id\":"); jb_emit_escaped(b, e->to_id ? e->to_id : ""); - jb_puts(b, ",\"relation\":"); jb_emit_escaped(b, e->relation ? e->relation : ""); - jb_puts(b, ",\"metadata\":"); jb_emit_escaped(b, e->metadata ? e->metadata : "{}"); - char tmp[64]; - snprintf(tmp, sizeof(tmp), ",\"weight\":%g", e->weight); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"confidence\":%g", e->confidence); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"created_at\":%lld", (long long)e->created_at); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"updated_at\":%lld", (long long)e->updated_at); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"last_fired\":%lld", (long long)e->last_fired); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"inhibitory\":%d", e->inhibitory ? 1 : 0); jb_puts(b, tmp); - snprintf(tmp, sizeof(tmp), ",\"layer_id\":%u", e->layer_id); jb_puts(b, tmp); - jb_putc(b, '}'); -} - -el_val_t engram_save(el_val_t path) { - const char* p = EL_CSTR(path); - if (!p || !*p) return 0; - EngramStore* g = engram_get(); - JsonBuf b; jb_init(&b); - jb_puts(&b, "{\"nodes\":["); - for (int64_t i = 0; i < g->node_count; i++) { - if (i > 0) jb_putc(&b, ','); - engram_emit_node_json(&b, &g->nodes[i]); - } - jb_puts(&b, "],\"edges\":["); - for (int64_t i = 0; i < g->edge_count; i++) { - if (i > 0) jb_putc(&b, ','); - engram_emit_edge_json(&b, &g->edges[i]); - } - /* Layered consciousness — emit the layer registry under "layers". - * Older readers that don't know about this top-level key will simply - * ignore it (forward compatible). Tombstoned (removed-injectable) - * layers are skipped — they have no name and can't be re-created - * meaningfully on load anyway. */ - jb_puts(&b, "],\"layers\":["); - int first_layer = 1; - for (size_t i = 0; i < g->layer_count; i++) { - EngramLayer* L = &g->layers[i]; - if (!L->name) continue; - if (!first_layer) jb_putc(&b, ','); - first_layer = 0; - jb_putc(&b, '{'); - char tmp[80]; - snprintf(tmp, sizeof(tmp), "\"layer_id\":%u", L->layer_id); - jb_puts(&b, tmp); - jb_puts(&b, ",\"name\":"); - jb_emit_escaped(&b, L->name); - snprintf(tmp, sizeof(tmp), ",\"activation_priority\":%u", L->activation_priority); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"suppressible\":%d", L->suppressible ? 1 : 0); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"transparent\":%d", L->transparent ? 1 : 0); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"injectable\":%d", L->injectable ? 1 : 0); - jb_puts(&b, tmp); - jb_putc(&b, '}'); - } - jb_puts(&b, "]}"); - FILE* f = fopen(p, "wb"); - if (!f) { free(b.buf); return 0; } - size_t w = fwrite(b.buf, 1, b.len, f); - fclose(f); - int ok = (w == b.len); - free(b.buf); - return ok ? 1 : 0; -} - -/* Helper: extract a string field from a JSON object substring. */ -static char* eg_get_str_field(const char* obj, const char* key) { - const char* p = json_find_key(obj, key); - if (!p) return el_strdup(""); - if (*p != '"') return el_strdup(""); - JsonParser jp = { .p = p, .end = p + strlen(p), .err = 0 }; - char* out = jp_parse_string_raw(&jp); - if (jp.err) { free(out); return el_strdup(""); } - return out; -} - -static double eg_get_num_field(const char* obj, const char* key) { - const char* p = json_find_key(obj, key); - if (!p || *p == '"' || *p == '{' || *p == '[') return 0.0; - return strtod(p, NULL); -} - -static int64_t eg_get_int_field(const char* obj, const char* key) { - const char* p = json_find_key(obj, key); - if (!p || *p == '"' || *p == '{' || *p == '[') return 0; - return strtoll(p, NULL, 10); -} - -/* Iterate the top-level nodes/edges arrays in a saved snapshot. */ -static const char* eg_skip_ws(const char* p) { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - return p; -} - -el_val_t engram_load(el_val_t path) { - const char* p = EL_CSTR(path); - if (!p || !*p) return 0; - FILE* f = fopen(p, "rb"); - if (!f) return 0; - fseek(f, 0, SEEK_END); - long sz = ftell(f); - rewind(f); - if (sz <= 0) { fclose(f); return 0; } - char* data = malloc((size_t)sz + 1); - if (!data) { fclose(f); return 0; } - size_t got = fread(data, 1, (size_t)sz, f); - fclose(f); - data[got] = '\0'; - - /* Reset store */ - EngramStore* g = engram_get(); - for (int64_t i = 0; i < g->node_count; i++) { - free(g->nodes[i].id); free(g->nodes[i].content); free(g->nodes[i].node_type); - free(g->nodes[i].label); free(g->nodes[i].tier); free(g->nodes[i].tags); - free(g->nodes[i].metadata); - } - g->node_count = 0; - for (int64_t i = 0; i < g->edge_count; i++) { - free(g->edges[i].id); free(g->edges[i].from_id); free(g->edges[i].to_id); - free(g->edges[i].relation); free(g->edges[i].metadata); - } - g->edge_count = 0; - - /* Walk nodes array */ - const char* nodes_p = json_find_key(data, "nodes"); - if (nodes_p) { - nodes_p = eg_skip_ws(nodes_p); - if (*nodes_p == '[') { - nodes_p++; - nodes_p = eg_skip_ws(nodes_p); - while (*nodes_p && *nodes_p != ']') { - if (*nodes_p != '{') { nodes_p++; continue; } - const char* end = json_skip_value(nodes_p); - size_t n = (size_t)(end - nodes_p); - char* obj = malloc(n + 1); - memcpy(obj, nodes_p, n); obj[n] = '\0'; - engram_grow_nodes(); - EngramNode* nn = &g->nodes[g->node_count]; - memset(nn, 0, sizeof(*nn)); - nn->id = eg_get_str_field(obj, "id"); - nn->content = eg_get_str_field(obj, "content"); - nn->node_type = eg_get_str_field(obj, "node_type"); - nn->label = eg_get_str_field(obj, "label"); - nn->tier = eg_get_str_field(obj, "tier"); - nn->tags = eg_get_str_field(obj, "tags"); - nn->metadata = eg_get_str_field(obj, "metadata"); - if (!nn->metadata || !*nn->metadata) { free(nn->metadata); nn->metadata = el_strdup("{}"); } - nn->salience = eg_get_num_field(obj, "salience"); - nn->importance = eg_get_num_field(obj, "importance"); - nn->confidence = eg_get_num_field(obj, "confidence"); - nn->temporal_decay_rate = eg_get_num_field(obj, "temporal_decay_rate"); - /* temporal_decay_rate defaults to 0 (use global) if absent in snapshot */ - nn->activation_count = eg_get_int_field(obj, "activation_count"); - nn->last_activated = eg_get_int_field(obj, "last_activated"); - nn->created_at = eg_get_int_field(obj, "created_at"); - nn->updated_at = eg_get_int_field(obj, "updated_at"); - nn->background_activation = eg_get_num_field(obj, "background_activation"); - nn->working_memory_weight = eg_get_num_field(obj, "working_memory_weight"); - nn->suppression_count = (int32_t)eg_get_int_field(obj, "suppression_count"); - /* layer_id defaults to ENGRAM_LAYER_DEFAULT (core-identity) - * for snapshots that predate the layered schema. We can't - * tell "explicit 0" from "missing field" using the helper - * directly, so probe for the key — if absent, fall back. */ - if (json_find_key(obj, "layer_id")) { - nn->layer_id = (uint32_t)eg_get_int_field(obj, "layer_id"); - } else { - nn->layer_id = ENGRAM_LAYER_DEFAULT; - } - g->node_count++; - free(obj); - nodes_p = end; - nodes_p = eg_skip_ws(nodes_p); - if (*nodes_p == ',') { nodes_p++; nodes_p = eg_skip_ws(nodes_p); } - } - } - } - /* Walk edges array */ - const char* edges_p = json_find_key(data, "edges"); - if (edges_p) { - edges_p = eg_skip_ws(edges_p); - if (*edges_p == '[') { - edges_p++; - edges_p = eg_skip_ws(edges_p); - while (*edges_p && *edges_p != ']') { - if (*edges_p != '{') { edges_p++; continue; } - const char* end = json_skip_value(edges_p); - size_t n = (size_t)(end - edges_p); - char* obj = malloc(n + 1); - memcpy(obj, edges_p, n); obj[n] = '\0'; - engram_grow_edges(); - EngramEdge* ee = &g->edges[g->edge_count]; - memset(ee, 0, sizeof(*ee)); - ee->id = eg_get_str_field(obj, "id"); - ee->from_id = eg_get_str_field(obj, "from_id"); - ee->to_id = eg_get_str_field(obj, "to_id"); - ee->relation = eg_get_str_field(obj, "relation"); - ee->metadata = eg_get_str_field(obj, "metadata"); - if (!ee->metadata || !*ee->metadata) { free(ee->metadata); ee->metadata = el_strdup("{}"); } - ee->weight = eg_get_num_field(obj, "weight"); - ee->confidence = eg_get_num_field(obj, "confidence"); - ee->created_at = eg_get_int_field(obj, "created_at"); - ee->updated_at = eg_get_int_field(obj, "updated_at"); - ee->last_fired = eg_get_int_field(obj, "last_fired"); - ee->inhibitory = (int)eg_get_int_field(obj, "inhibitory"); - if (json_find_key(obj, "layer_id")) { - ee->layer_id = (uint32_t)eg_get_int_field(obj, "layer_id"); - } else { - ee->layer_id = ENGRAM_LAYER_DEFAULT; - } - g->edge_count++; - free(obj); - edges_p = end; - edges_p = eg_skip_ws(edges_p); - if (*edges_p == ',') { edges_p++; edges_p = eg_skip_ws(edges_p); } - } - } - } - /* Walk layers array (optional — older snapshots omit this). - * If present we replace the canonical registry entirely; if absent we - * keep whatever the engram_get() init established. */ - const char* layers_p = json_find_key(data, "layers"); - if (layers_p) { - layers_p = eg_skip_ws(layers_p); - if (*layers_p == '[') { - /* Reset existing layer registry. Free strdup'd names; the - * struct array itself can be reused. */ - for (size_t i = 0; i < g->layer_count; i++) { - if (g->layers[i].name) free(g->layers[i].name); - g->layers[i].name = NULL; - } - g->layer_count = 0; - - layers_p++; - layers_p = eg_skip_ws(layers_p); - while (*layers_p && *layers_p != ']') { - if (*layers_p != '{') { layers_p++; continue; } - const char* end = json_skip_value(layers_p); - size_t n = (size_t)(end - layers_p); - char* obj = malloc(n + 1); - memcpy(obj, layers_p, n); obj[n] = '\0'; - if (g->layer_count >= g->layer_capacity) { - size_t nc = g->layer_capacity ? g->layer_capacity * 2 : 16; - EngramLayer* grown = realloc(g->layers, nc * sizeof(EngramLayer)); - if (!grown) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - memset(grown + g->layer_capacity, 0, - (nc - g->layer_capacity) * sizeof(EngramLayer)); - g->layers = grown; - g->layer_capacity = nc; - } - EngramLayer* L = &g->layers[g->layer_count]; - memset(L, 0, sizeof(*L)); - L->layer_id = (uint32_t)eg_get_int_field(obj, "layer_id"); - L->activation_priority = (uint32_t)eg_get_int_field(obj, "activation_priority"); - L->suppressible = (int)eg_get_int_field(obj, "suppressible") ? 1 : 0; - L->transparent = (int)eg_get_int_field(obj, "transparent") ? 1 : 0; - L->injectable = (int)eg_get_int_field(obj, "injectable") ? 1 : 0; - char* nm = eg_get_str_field(obj, "name"); - if (nm && *nm) { - L->name = el_strdup_persist(nm); - free(nm); - } else { - free(nm); - L->name = el_strdup_persist(""); - } - g->layer_count++; - free(obj); - layers_p = end; - layers_p = eg_skip_ws(layers_p); - if (*layers_p == ',') { layers_p++; layers_p = eg_skip_ws(layers_p); } - } - } - } - free(data); - return 1; -} - -/* ── Engram JSON-string accessors ───────────────────────────────────────── - * These return pre-serialized JSON strings so callers (especially HTTP - * handlers) don't have to round-trip ElList/ElMap through json_stringify - * — which can't reliably distinguish those structures from raw pointers - * due to el_val_t's type erasure. The runtime knows the real C types and - * can serialize directly. */ - -el_val_t engram_get_node_json(el_val_t id) { - const char* sid = EL_CSTR(id); - EngramNode* n = engram_find_node(sid); - if (!n) return el_wrap_str(el_strdup("{}")); - JsonBuf b; jb_init(&b); - engram_emit_node_json(&b, n); - return el_wrap_str(b.buf); -} - -el_val_t engram_search_json(el_val_t query, el_val_t limit) { - EngramStore* g = engram_get(); - const char* q = EL_CSTR(query); - int64_t lim = (int64_t)limit; - if (lim <= 0) lim = 100; - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - int first = 1; - int64_t found = 0; - if (q && *q) { - for (int64_t i = 0; i < g->node_count && found < lim; i++) { - EngramNode* n = &g->nodes[i]; - /* Filter transparent layers — same as engram_search. */ - if (engram_layer_is_transparent(n->layer_id)) continue; - if (istr_contains(n->content, q) || - istr_contains(n->label, q) || - istr_contains(n->tags, q)) { - if (!first) jb_putc(&b, ','); - engram_emit_node_json(&b, n); - first = 0; - found++; - } - } - } - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -el_val_t engram_scan_nodes_json(el_val_t limit, el_val_t offset) { - EngramStore* g = engram_get(); - int64_t lim = (int64_t)limit; if (lim <= 0) lim = 100; - int64_t off = (int64_t)offset; if (off < 0) off = 0; - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - if (g->node_count == 0) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - int64_t* idx = malloc((size_t)g->node_count * sizeof(int64_t)); - if (!idx) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - /* Skip transparent layers — introspection filter, same as engram_scan_nodes. */ - int64_t live = 0; - for (int64_t i = 0; i < g->node_count; i++) { - if (engram_layer_is_transparent(g->nodes[i].layer_id)) continue; - idx[live++] = i; - } - engram_sort_indices_by_salience(idx, live, g->nodes); - int64_t end = off + lim; - if (end > live) end = live; - int first = 1; - for (int64_t i = off; i < end; i++) { - if (!first) jb_putc(&b, ','); - engram_emit_node_json(&b, &g->nodes[idx[i]]); - first = 0; - } - free(idx); - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -/* engram_scan_nodes_by_type_json — filter by node_type before paginating. - * Empty / NULL type_v falls back to the unfiltered scan (existing behaviour). - * Result is JSON array, salience-sorted, transparent layers skipped. */ -el_val_t engram_scan_nodes_by_type_json(el_val_t type_v, el_val_t limit, el_val_t offset) { - const char* type_filter = EL_CSTR(type_v); - if (!type_filter || !*type_filter) { - return engram_scan_nodes_json(limit, offset); - } - EngramStore* g = engram_get(); - int64_t lim = (int64_t)limit; if (lim <= 0) lim = 100; - int64_t off = (int64_t)offset; if (off < 0) off = 0; - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - if (g->node_count == 0) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - int64_t* idx = malloc((size_t)g->node_count * sizeof(int64_t)); - if (!idx) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - int64_t live = 0; - for (int64_t i = 0; i < g->node_count; i++) { - if (engram_layer_is_transparent(g->nodes[i].layer_id)) continue; - const char* nt = g->nodes[i].node_type; - if (!nt || strcmp(nt, type_filter) != 0) continue; - idx[live++] = i; - } - engram_sort_indices_by_salience(idx, live, g->nodes); - int64_t end = off + lim; - if (end > live) end = live; - int first = 1; - for (int64_t i = off; i < end; i++) { - if (!first) jb_putc(&b, ','); - engram_emit_node_json(&b, &g->nodes[idx[i]]); - first = 0; - } - free(idx); - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -el_val_t engram_neighbors_json(el_val_t node_id, el_val_t max_depth, el_val_t direction) { - /* Re-implement here directly so we serialize without going through - * the ElList path. Walks BFS to max_depth, emits {node, edge, hops} - * triples. */ - EngramStore* g = engram_get(); - const char* sid = EL_CSTR(node_id); - int64_t depth = (int64_t)max_depth; if (depth <= 0) depth = 1; - const char* dir = EL_CSTR(direction); if (!dir) dir = "both"; - int allow_out = (strcmp(dir, "out") == 0) || (strcmp(dir, "both") == 0); - int allow_in = (strcmp(dir, "in") == 0) || (strcmp(dir, "both") == 0); - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - if (!sid || !*sid) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - - /* Frontier of (node_id, hops). Cap to a sane size. */ - char** frontier = calloc(1024, sizeof(char*)); - int64_t* frontier_h = calloc(1024, sizeof(int64_t)); - int64_t fc = 0; - char** visited = calloc(1024, sizeof(char*)); - int64_t vc = 0; - if (!frontier || !frontier_h || !visited) { - free(frontier); free(frontier_h); free(visited); - jb_putc(&b, ']'); return el_wrap_str(b.buf); - } - frontier[fc] = el_strdup(sid); frontier_h[fc] = 0; fc++; - visited[vc++] = el_strdup(sid); - - int first = 1; - while (fc > 0) { - char* cur = frontier[0]; int64_t h = frontier_h[0]; - for (int64_t k = 1; k < fc; k++) { frontier[k-1] = frontier[k]; frontier_h[k-1] = frontier_h[k]; } - fc--; - if (h >= depth) { free(cur); continue; } - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - const char* peer = NULL; - if (allow_out && e->from_id && strcmp(e->from_id, cur) == 0) peer = e->to_id; - else if (allow_in && e->to_id && strcmp(e->to_id, cur) == 0) peer = e->from_id; - if (!peer) continue; - int seen = 0; - for (int64_t v = 0; v < vc; v++) { - if (strcmp(visited[v], peer) == 0) { seen = 1; break; } - } - if (seen) continue; - EngramNode* n = engram_find_node(peer); - if (!n) continue; - if (!first) jb_putc(&b, ','); - jb_puts(&b, "{\"node\":"); - engram_emit_node_json(&b, n); - jb_puts(&b, ",\"edge\":"); - engram_emit_edge_json(&b, e); - char tmp[64]; snprintf(tmp, sizeof(tmp), ",\"hops\":%lld}", (long long)(h + 1)); - jb_puts(&b, tmp); - first = 0; - if (vc < 1024) visited[vc++] = el_strdup(peer); - if (fc < 1024 && h + 1 < depth) { frontier[fc] = el_strdup(peer); frontier_h[fc] = h + 1; fc++; } - } - free(cur); - } - for (int64_t i = 0; i < fc; i++) free(frontier[i]); - for (int64_t i = 0; i < vc; i++) free(visited[i]); - free(frontier); free(frontier_h); free(visited); - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -el_val_t engram_activate_json(el_val_t query, el_val_t depth) { - /* Run two-layer engram_activate and serialize the result list to JSON. - * Each entry includes both activation_strength (layer 1 background) and - * working_memory_weight (layer 2 executive filter), plus promoted flag. - * Callers performing context compilation should filter to promoted=1. */ - el_val_t lst = engram_activate(query, depth); - ElList* arr = (ElList*)(uintptr_t)lst; - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - if (arr) { - for (int64_t i = 0; i < arr->length; i++) { - if (!arr->elems[i]) continue; - el_val_t node_map = el_map_get(arr->elems[i], EL_STR("node")); - el_val_t strength_v = el_map_get(arr->elems[i], EL_STR("activation_strength")); - el_val_t wm_v = el_map_get(arr->elems[i], EL_STR("working_memory_weight")); - el_val_t epist_v = el_map_get(arr->elems[i], EL_STR("epistemic_confidence")); - el_val_t hops_v = el_map_get(arr->elems[i], EL_STR("hops")); - el_val_t promoted_v = el_map_get(arr->elems[i], EL_STR("promoted")); - /* Look up underlying EngramNode by id to emit canonical JSON. */ - el_val_t id_v = el_map_get(node_map, EL_STR("id")); - const char* id_s = EL_CSTR(id_v); - EngramNode* n = id_s ? engram_find_node(id_s) : NULL; - if (i > 0) jb_putc(&b, ','); - jb_puts(&b, "{\"node\":"); - if (n) { - engram_emit_node_json(&b, n); - } else { - jb_puts(&b, "{}"); - } - char tmp[80]; - snprintf(tmp, sizeof(tmp), ",\"activation_strength\":%g", el_to_float(strength_v)); jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"working_memory_weight\":%g", el_to_float(wm_v)); jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"epistemic_confidence\":%g", el_to_float(epist_v)); jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"hops\":%lld", (long long)(int64_t)hops_v); jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"promoted\":%d}", (int)(int64_t)promoted_v); jb_puts(&b, tmp); - } - } - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -el_val_t engram_stats_json(void) { - EngramStore* g = engram_get(); - char buf[128]; - snprintf(buf, sizeof(buf), - "{\"node_count\":%lld,\"edge_count\":%lld,\"layer_count\":%zu}", - (long long)g->node_count, (long long)g->edge_count, g->layer_count); - return el_wrap_str(el_strdup(buf)); -} - -/* engram_list_layers_json — serialized counterpart of engram_list_layers. - * Returns a JSON array, sorted by activation_priority ascending. */ -el_val_t engram_list_layers_json(void) { - EngramStore* g = engram_get(); - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - /* Build a sorted index over live layers. */ - size_t* idx = malloc((g->layer_count + 1) * sizeof(size_t)); - if (!idx) { jb_putc(&b, ']'); return el_wrap_str(b.buf); } - size_t live = 0; - for (size_t i = 0; i < g->layer_count; i++) { - if (g->layers[i].name) idx[live++] = i; - } - for (size_t i = 1; i < live; i++) { - size_t key = idx[i]; - uint32_t kp = g->layers[key].activation_priority; - size_t j = i; - while (j > 0 && g->layers[idx[j - 1]].activation_priority > kp) { - idx[j] = idx[j - 1]; - j--; - } - idx[j] = key; - } - int first = 1; - for (size_t i = 0; i < live; i++) { - EngramLayer* L = &g->layers[idx[i]]; - if (!first) jb_putc(&b, ','); - first = 0; - jb_putc(&b, '{'); - char tmp[80]; - snprintf(tmp, sizeof(tmp), "\"layer_id\":%u", L->layer_id); jb_puts(&b, tmp); - jb_puts(&b, ",\"name\":"); - jb_emit_escaped(&b, L->name ? L->name : ""); - snprintf(tmp, sizeof(tmp), ",\"activation_priority\":%u", L->activation_priority); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"suppressible\":%d", L->suppressible ? 1 : 0); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"transparent\":%d", L->transparent ? 1 : 0); - jb_puts(&b, tmp); - snprintf(tmp, sizeof(tmp), ",\"injectable\":%d", L->injectable ? 1 : 0); - jb_puts(&b, tmp); - jb_putc(&b, '}'); - } - free(idx); - jb_putc(&b, ']'); - return el_wrap_str(b.buf); -} - -/* engram_compile_layered_json — produce a prompt-ready context block split - * by layer. - * - * Runs the three-pass activation, then partitions promoted nodes by layer - * suppressibility: - * - Non-suppressible (Layer 0 / structural-floor) layers go FIRST under - * the heading "[LAYER 0 — STRUCTURAL]". These are the sacred-fire - * nodes that surfaced via the pass-3 override. - * - All other promoted layers go SECOND under "[ENGRAM CONTEXT]". - * - * Output is a single JSON-string el_val_t: a UTF-8 text block ready to be - * concatenated into a system prompt. Returns "" if no nodes promoted. - * - * Transparent layers (Layer 0) are emitted into the prompt — they shape - * the model's output — but engram_search and friends still hide them from - * introspection-style queries. The split heading lets the LLM weight them - * appropriately without revealing their internal label. - * - * Each emitted line for a node is its raw JSON (matching engram_emit_node_json) - * so downstream JSON parsers can still walk individual records inside the - * formatted block. The block is plain text, not a JSON document — callers - * concatenating it into a prompt should treat it as opaque markdown. */ -el_val_t engram_compile_layered_json(el_val_t intent, el_val_t depth) { - EngramStore* g = engram_get(); - /* Run the three-pass activator. We need the persisted node fields, so - * call engram_activate (it writes background_activation and - * working_memory_weight back into the store). */ - (void)engram_activate(intent, depth); - - /* Walk the store and partition by suppressibility. */ - JsonBuf b; jb_init(&b); - int wrote_layer0 = 0; - int wrote_normal = 0; - - /* Sort indices by working_memory_weight descending so the most - * confidently promoted nodes appear first within each section. */ - int64_t* idx = malloc((size_t)(g->node_count + 1) * sizeof(int64_t)); - if (!idx) return el_wrap_str(el_strdup("")); - int64_t mc = 0; - for (int64_t i = 0; i < g->node_count; i++) { - if (g->nodes[i].working_memory_weight > 0.0) idx[mc++] = i; - } - for (int64_t i = 1; i < mc; i++) { - int64_t key = idx[i]; - double kw = g->nodes[key].working_memory_weight; - int64_t j = i; - while (j > 0 && g->nodes[idx[j - 1]].working_memory_weight < kw) { - idx[j] = idx[j - 1]; - j--; - } - idx[j] = key; - } - - /* Section 1: structural floor (non-suppressible layers). */ - for (int64_t i = 0; i < mc; i++) { - EngramNode* n = &g->nodes[idx[i]]; - if (engram_layer_is_suppressible(n->layer_id)) continue; - if (!wrote_layer0) { - jb_puts(&b, "[LAYER 0 — STRUCTURAL]\n"); - wrote_layer0 = 1; - } - engram_emit_node_json(&b, n); - jb_putc(&b, '\n'); - } - - /* Section 2: standard engram context (suppressible layers). */ - for (int64_t i = 0; i < mc; i++) { - EngramNode* n = &g->nodes[idx[i]]; - if (!engram_layer_is_suppressible(n->layer_id)) continue; - if (!wrote_normal) { - if (wrote_layer0) jb_putc(&b, '\n'); - jb_puts(&b, "[ENGRAM CONTEXT]\n"); - wrote_normal = 1; - } - engram_emit_node_json(&b, n); - jb_putc(&b, '\n'); - } - - free(idx); - if (b.len == 0) { - free(b.buf); - return el_wrap_str(el_strdup("")); - } - return el_wrap_str(b.buf); -} - -/* engram_query_range — temporal range query. - * Returns a JSON array of nodes whose created_at OR last_activated falls - * within [start_ms, end_ms], sorted by created_at ascending. - * Enables "what was I working on last Tuesday?" style queries by passing - * unix-millisecond timestamps for the start and end of the target interval. - * Both endpoints are inclusive. Pass 0 for start_ms to mean "beginning of - * time"; pass 0 for end_ms to mean "now". */ -el_val_t engram_query_range(el_val_t start_ms_v, el_val_t end_ms_v) { - EngramStore* g = engram_get(); - int64_t start_ms = (int64_t)start_ms_v; - int64_t end_ms = (int64_t)end_ms_v; - if (end_ms <= 0) end_ms = engram_now_ms(); - - /* Collect matching indices. */ - int64_t* idx = malloc((size_t)g->node_count * sizeof(int64_t)); - if (!idx) return el_wrap_str(el_strdup("[]")); - int64_t mc = 0; - for (int64_t i = 0; i < g->node_count; i++) { - EngramNode* n = &g->nodes[i]; - int in_created = (n->created_at >= start_ms && n->created_at <= end_ms); - int in_activated = (n->last_activated >= start_ms && n->last_activated <= end_ms); - if (in_created || in_activated) idx[mc++] = i; - } - /* Sort by created_at ascending (insertion sort — N is small in practice). */ - for (int64_t i = 1; i < mc; i++) { - int64_t key = idx[i]; - int64_t kts = g->nodes[key].created_at; - int64_t j = i - 1; - while (j >= 0 && g->nodes[idx[j]].created_at > kts) { - idx[j + 1] = idx[j]; - j--; - } - idx[j + 1] = key; - } - JsonBuf b; jb_init(&b); - jb_putc(&b, '['); - for (int64_t i = 0; i < mc; i++) { - if (i > 0) jb_putc(&b, ','); - engram_emit_node_json(&b, &g->nodes[idx[i]]); - } - jb_putc(&b, ']'); - free(idx); - return el_wrap_str(b.buf); -} - -/* ── DHARMA network ───────────────────────────────────────────────────────── - * Real implementation. Peers are addressed by `dharma_id` — either bare - * (e.g. "ntn-genesis", transport defaults to http://localhost:7770) or - * "@" where is the peer's Engram-exposed daemon. - * - * Channels are logical handles cached per-cgi: `dharma_connect` is - * idempotent and returns "ch:". The channel registry below tracks - * every cgi_id we've connected to and its resolved transport URL. - * - * Relationship weights live in the local Engram graph: edges of type - * "dharma-relation" between a synthetic local node ("dharma:self") and - * synthetic peer nodes ("dharma:peer:"). Hebbian increments - * accumulate in EngramEdge.weight, clamped to [0.0, 1.0]. - * - * Events arrive over HTTP via the application's request handler, which is - * expected to call el_runtime_dharma_event_arrive() when it sees a - * /dharma/event POST. dharma_field() blocks on a per-event-type queue. - */ - -#define DHARMA_DEFAULT_URL "http://localhost:7770" - -/* Channel registry — one entry per known peer. */ -typedef struct DharmaChannel { - char* cgi_id; /* full dharma_id including any @ suffix */ - char* base_id; /* registry-id portion (before @) for relationship lookup */ - char* url; /* resolved transport URL */ - char* channel_id; /* "ch:" */ -} DharmaChannel; - -static DharmaChannel* _dharma_channels = NULL; -static size_t _dharma_channel_count = 0; -static size_t _dharma_channel_cap = 0; -static pthread_mutex_t _dharma_channel_mu = PTHREAD_MUTEX_INITIALIZER; - -/* Event queue — per-type linked list. dharma_field blocks on _dharma_event_cv. */ -typedef struct DharmaEvent { - char* event_type; - char* payload; - char* source; - int64_t timestamp; - struct DharmaEvent* next; -} DharmaEvent; - -static DharmaEvent* _dharma_event_head = NULL; -static DharmaEvent* _dharma_event_tail = NULL; -static pthread_mutex_t _dharma_event_mu = PTHREAD_MUTEX_INITIALIZER; -static pthread_cond_t _dharma_event_cv = PTHREAD_COND_INITIALIZER; - -/* Split "@" → (base_id, url). If no "@", base_id = full, url = default. - * Returned strings are heap-allocated; caller must free. */ -static void dharma_parse_id(const char* full, char** out_base, char** out_url) { - if (!full) full = ""; - const char* at = strchr(full, '@'); - if (at) { - size_t bn = (size_t)(at - full); - char* b = malloc(bn + 1); - memcpy(b, full, bn); b[bn] = '\0'; - *out_base = b; - *out_url = el_strdup(at + 1); - if (!**out_url) { free(*out_url); *out_url = el_strdup(DHARMA_DEFAULT_URL); } - } else { - *out_base = el_strdup(full); - *out_url = el_strdup(DHARMA_DEFAULT_URL); - } -} - -/* Find existing channel by full cgi_id. Caller must hold _dharma_channel_mu. */ -static DharmaChannel* dharma_find_channel_locked(const char* cgi_id) { - if (!cgi_id) return NULL; - for (size_t i = 0; i < _dharma_channel_count; i++) { - if (_dharma_channels[i].cgi_id && - strcmp(_dharma_channels[i].cgi_id, cgi_id) == 0) { - return &_dharma_channels[i]; - } - } - return NULL; -} - -/* Add a new channel entry. Caller must hold _dharma_channel_mu. */ -static DharmaChannel* dharma_add_channel_locked(const char* cgi_id) { - if (_dharma_channel_count >= _dharma_channel_cap) { - size_t nc = _dharma_channel_cap ? _dharma_channel_cap * 2 : 8; - _dharma_channels = realloc(_dharma_channels, nc * sizeof(DharmaChannel)); - if (!_dharma_channels) { fputs("el_runtime: out of memory\n", stderr); exit(1); } - memset(_dharma_channels + _dharma_channel_cap, 0, - (nc - _dharma_channel_cap) * sizeof(DharmaChannel)); - _dharma_channel_cap = nc; - } - DharmaChannel* ch = &_dharma_channels[_dharma_channel_count++]; - char* base = NULL; char* url = NULL; - dharma_parse_id(cgi_id, &base, &url); - ch->cgi_id = el_strdup(cgi_id ? cgi_id : ""); - ch->base_id = base; - ch->url = url; - size_t cn = strlen(ch->cgi_id) + 4; - ch->channel_id = malloc(cn); - snprintf(ch->channel_id, cn, "ch:%s", ch->cgi_id); - return ch; -} - -el_val_t dharma_connect(el_val_t cgi_id) { - const char* id = EL_CSTR(cgi_id); - if (!id || !*id) return el_wrap_str(el_strdup("")); - pthread_mutex_lock(&_dharma_channel_mu); - DharmaChannel* ch = dharma_find_channel_locked(id); - if (!ch) ch = dharma_add_channel_locked(id); - char* out = el_strdup(ch->channel_id); - pthread_mutex_unlock(&_dharma_channel_mu); - return el_wrap_str(out); -} - -/* Build an error JSON body — same shape http_error_json uses. */ -static el_val_t dharma_error_json(const char* msg) { - return http_error_json(msg); -} - -el_val_t dharma_send(el_val_t channel, el_val_t content) { - const char* ch_id = EL_CSTR(channel); - const char* msg = EL_CSTR(content); - if (!ch_id || strncmp(ch_id, "ch:", 3) != 0) { - return dharma_error_json("invalid channel"); - } - const char* peer_id = ch_id + 3; - /* Look up channel; if unknown (caller fabricated), auto-register. */ - pthread_mutex_lock(&_dharma_channel_mu); - DharmaChannel* ch = dharma_find_channel_locked(peer_id); - if (!ch) ch = dharma_add_channel_locked(peer_id); - char* url = el_strdup(ch->url); - pthread_mutex_unlock(&_dharma_channel_mu); - /* Build /dharma/recv body. */ - const char* from = _el_cgi_dharma_id ? _el_cgi_dharma_id : "(unknown)"; - char* esc_ch = json_escape_alloc(ch_id); - char* esc_from = json_escape_alloc(from); - char* esc_msg = json_escape_alloc(msg ? msg : ""); - JsonBuf b; jb_init(&b); - jb_puts(&b, "{\"channel\":\""); jb_puts(&b, esc_ch); - jb_puts(&b, "\",\"from\":\""); jb_puts(&b, esc_from); - jb_puts(&b, "\",\"content\":\""); jb_puts(&b, esc_msg); - jb_puts(&b, "\"}"); - free(esc_ch); free(esc_from); free(esc_msg); - size_t ul = strlen(url) + 16; - char* full_url = malloc(ul); - snprintf(full_url, ul, "%s/dharma/recv", url); - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - el_val_t resp = http_do("POST", full_url, b.buf, h); - curl_slist_free_all(h); - free(b.buf); free(full_url); free(url); - return resp; -} - -el_val_t dharma_activate(el_val_t query) { - const char* q = EL_CSTR(query); - if (!q) q = ""; - el_val_t out = el_list_empty(); - char* esc_q = json_escape_alloc(q); - JsonBuf body; jb_init(&body); - jb_puts(&body, "{\"query\":\""); jb_puts(&body, esc_q); jb_puts(&body, "\"}"); - free(esc_q); - - /* Snapshot the channel list under lock so we can iterate without - * holding the mutex during network I/O. */ - pthread_mutex_lock(&_dharma_channel_mu); - size_t n = _dharma_channel_count; - char** urls = calloc(n ? n : 1, sizeof(char*)); - char** ids = calloc(n ? n : 1, sizeof(char*)); - char** bases = calloc(n ? n : 1, sizeof(char*)); - for (size_t i = 0; i < n; i++) { - urls[i] = el_strdup(_dharma_channels[i].url); - ids[i] = el_strdup(_dharma_channels[i].cgi_id); - bases[i] = el_strdup(_dharma_channels[i].base_id); - } - pthread_mutex_unlock(&_dharma_channel_mu); - - for (size_t i = 0; i < n; i++) { - size_t ul = strlen(urls[i]) + 32; - char* full_url = malloc(ul); - snprintf(full_url, ul, "%s/api/activate", urls[i]); - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - el_val_t resp = http_do("POST", full_url, body.buf, h); - curl_slist_free_all(h); - free(full_url); - const char* rs = EL_CSTR(resp); - if (!rs || !*rs) continue; - if (rs[0] == '{' && strstr(rs, "\"error\"")) continue; - - /* Look up relationship weight (attenuation). */ - double rel_weight = 1.0; - { - const char* self_id = "dharma:self"; - char peer_node[512]; - snprintf(peer_node, sizeof(peer_node), "dharma:peer:%s", bases[i]); - EngramStore* g = engram_get(); - for (int64_t k = 0; k < g->edge_count; k++) { - EngramEdge* e = &g->edges[k]; - if (e->from_id && e->to_id && - strcmp(e->from_id, self_id) == 0 && - strcmp(e->to_id, peer_node) == 0 && - e->relation && strcmp(e->relation, "dharma-relation") == 0) { - rel_weight = e->weight; - break; - } - } - } - - /* Iterate the response array. Expect either a top-level array - * or an object whose "results" field is an array. */ - const char* arr = rs; - while (*arr == ' ' || *arr == '\t' || *arr == '\n' || *arr == '\r') arr++; - char* arr_owned = NULL; - if (*arr == '{') { - el_val_t r = json_get_raw(EL_STR(rs), EL_STR("results")); - const char* rr = EL_CSTR(r); - if (rr && *rr == '[') { - arr_owned = el_strdup(rr); - arr = arr_owned; - } else { - continue; - } - } - if (*arr != '[') { free(arr_owned); continue; } - const char* p = arr + 1; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - while (*p && *p != ']') { - const char* end = json_skip_value(p); - size_t en = (size_t)(end - p); - char* obj = el_strbuf(en); - memcpy(obj, p, en); obj[en] = '\0'; - - /* Pull activation_strength if present, else 1.0. */ - el_val_t act_v = json_get_float(EL_STR(obj), EL_STR("activation_strength")); - double act = el_to_float(act_v); - if (!(act > 0.0 && act <= 100.0)) act = 1.0; - double final_act = act * rel_weight; - - el_val_t entry = el_map_new(0); - /* node = the inner JSON if present, else the entire obj. */ - el_val_t node_raw = json_get_raw(EL_STR(obj), EL_STR("node")); - const char* nr = EL_CSTR(node_raw); - entry = el_map_set(entry, EL_STR(el_strdup("node")), - (nr && *nr) ? node_raw : EL_STR(el_strdup(obj))); - entry = el_map_set(entry, EL_STR(el_strdup("source_cgi")), - EL_STR(el_strdup(ids[i]))); - entry = el_map_set(entry, EL_STR(el_strdup("activation_strength")), - el_from_float(final_act)); - out = el_list_append(out, entry); - free(obj); - p = end; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - } - free(arr_owned); - } - for (size_t i = 0; i < n; i++) { free(urls[i]); free(ids[i]); free(bases[i]); } - free(urls); free(ids); free(bases); - free(body.buf); - return out; -} - -void dharma_emit(el_val_t event_type, el_val_t payload) { - const char* et = EL_CSTR(event_type); - const char* pay = EL_CSTR(payload); - if (!et) et = ""; - if (!pay) pay = ""; - const char* src = _el_cgi_dharma_id ? _el_cgi_dharma_id : "(unknown)"; - int64_t ts = engram_now_ms(); - - char* esc_et = json_escape_alloc(et); - char* esc_pay = json_escape_alloc(pay); - char* esc_src = json_escape_alloc(src); - JsonBuf b; jb_init(&b); - jb_puts(&b, "{\"type\":\""); jb_puts(&b, esc_et); - jb_puts(&b, "\",\"payload\":\""); jb_puts(&b, esc_pay); - jb_puts(&b, "\",\"source\":\""); jb_puts(&b, esc_src); - jb_puts(&b, "\",\"timestamp\":"); jb_emit_int(&b, ts); - jb_putc(&b, '}'); - free(esc_et); free(esc_pay); free(esc_src); - - /* Snapshot URLs to avoid holding the channel mutex during I/O. */ - pthread_mutex_lock(&_dharma_channel_mu); - size_t n = _dharma_channel_count; - char** urls = calloc(n ? n : 1, sizeof(char*)); - for (size_t i = 0; i < n; i++) urls[i] = el_strdup(_dharma_channels[i].url); - pthread_mutex_unlock(&_dharma_channel_mu); - - for (size_t i = 0; i < n; i++) { - size_t ul = strlen(urls[i]) + 32; - char* full_url = malloc(ul); - snprintf(full_url, ul, "%s/dharma/event", urls[i]); - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - el_val_t r = http_do("POST", full_url, b.buf, h); - (void)r; /* fire-and-forget — emit is not synchronous */ - curl_slist_free_all(h); - free(full_url); - } - for (size_t i = 0; i < n; i++) free(urls[i]); - free(urls); - free(b.buf); -} - -void el_runtime_dharma_event_arrive(const char* event_type, const char* payload, - const char* source) { - DharmaEvent* ev = calloc(1, sizeof(DharmaEvent)); - if (!ev) return; - ev->event_type = el_strdup(event_type ? event_type : ""); - ev->payload = el_strdup(payload ? payload : ""); - ev->source = el_strdup(source ? source : ""); - ev->timestamp = engram_now_ms(); - ev->next = NULL; - pthread_mutex_lock(&_dharma_event_mu); - if (_dharma_event_tail) _dharma_event_tail->next = ev; - else _dharma_event_head = ev; - _dharma_event_tail = ev; - pthread_cond_broadcast(&_dharma_event_cv); - pthread_mutex_unlock(&_dharma_event_mu); -} - -el_val_t dharma_field(el_val_t event_type) { - const char* et = EL_CSTR(event_type); - if (!et) et = ""; - - /* Compute deadline: now + 30 seconds. */ - struct timespec deadline; - clock_gettime(CLOCK_REALTIME, &deadline); - deadline.tv_sec += 30; - - DharmaEvent* found = NULL; - pthread_mutex_lock(&_dharma_event_mu); - while (1) { - /* Scan queue for matching type; pop and return first match. */ - DharmaEvent* prev = NULL; - DharmaEvent* cur = _dharma_event_head; - while (cur) { - if (cur->event_type && strcmp(cur->event_type, et) == 0) { - if (prev) prev->next = cur->next; - else _dharma_event_head = cur->next; - if (_dharma_event_tail == cur) _dharma_event_tail = prev; - cur->next = NULL; - found = cur; - break; - } - prev = cur; cur = cur->next; - } - if (found) break; - int rc = pthread_cond_timedwait(&_dharma_event_cv, &_dharma_event_mu, &deadline); - if (rc == ETIMEDOUT) break; - } - pthread_mutex_unlock(&_dharma_event_mu); - - if (!found) return el_map_new(0); - el_val_t m = el_map_new(0); - m = el_map_set(m, EL_STR(el_strdup("type")), - EL_STR(el_strdup(found->event_type ? found->event_type : ""))); - m = el_map_set(m, EL_STR(el_strdup("payload")), - EL_STR(el_strdup(found->payload ? found->payload : ""))); - m = el_map_set(m, EL_STR(el_strdup("source_cgi")), - EL_STR(el_strdup(found->source ? found->source : ""))); - m = el_map_set(m, EL_STR(el_strdup("timestamp")), (el_val_t)found->timestamp); - free(found->event_type); free(found->payload); free(found->source); free(found); - return m; -} - -/* Locate (or create) the local "dharma:self" node and the synthetic peer - * node "dharma:peer:". Returns the index of the dharma-relation - * edge, or -1 if not found. If `create` is non-zero, ensure the nodes - * and edge exist (creating them as needed) and return the edge index. */ -static int64_t dharma_find_or_create_relation_edge(const char* peer_base, int create) { - if (!peer_base || !*peer_base) return -1; - EngramStore* g = engram_get(); - const char* self_id = "dharma:self"; - char peer_node[512]; - snprintf(peer_node, sizeof(peer_node), "dharma:peer:%s", peer_base); - - /* Look for the edge first. */ - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - if (e->from_id && e->to_id && - strcmp(e->from_id, self_id) == 0 && - strcmp(e->to_id, peer_node) == 0 && - e->relation && strcmp(e->relation, "dharma-relation") == 0) { - return i; - } - } - if (!create) return -1; - - /* Ensure self node exists. We use a fixed id (not engram_new_id) so - * subsequent calls reuse the same one. */ - if (!engram_find_node(self_id)) { - engram_grow_nodes(); - EngramNode* n = &g->nodes[g->node_count]; - memset(n, 0, sizeof(*n)); - n->id = el_strdup(self_id); - n->content = el_strdup(_el_cgi_dharma_id ? _el_cgi_dharma_id : "(self)"); - n->node_type = el_strdup("DharmaSelf"); - n->label = el_strdup("dharma:self"); - n->tier = el_strdup("Working"); - n->tags = el_strdup("dharma"); - n->metadata = el_strdup("{}"); - n->salience = 1.0; n->importance = 1.0; n->confidence = 1.0; - int64_t now = engram_now_ms(); - n->created_at = now; n->updated_at = now; n->last_activated = now; - n->layer_id = ENGRAM_LAYER_DEFAULT; - g->node_count++; - } - if (!engram_find_node(peer_node)) { - engram_grow_nodes(); - EngramNode* n = &g->nodes[g->node_count]; - memset(n, 0, sizeof(*n)); - n->id = el_strdup(peer_node); - n->content = el_strdup(peer_base); - n->node_type = el_strdup("DharmaPeer"); - n->label = el_strdup(peer_node); - n->tier = el_strdup("Working"); - n->tags = el_strdup("dharma"); - n->metadata = el_strdup("{}"); - n->salience = 0.5; n->importance = 0.5; n->confidence = 1.0; - int64_t now = engram_now_ms(); - n->created_at = now; n->updated_at = now; n->last_activated = now; - n->layer_id = ENGRAM_LAYER_DEFAULT; - g->node_count++; - } - /* Create the edge with weight 0.0 — caller will increment. */ - engram_grow_edges(); - EngramEdge* e = &g->edges[g->edge_count]; - memset(e, 0, sizeof(*e)); - e->id = engram_new_id(); - e->from_id = el_strdup(self_id); - e->to_id = el_strdup(peer_node); - e->relation = el_strdup("dharma-relation"); - e->metadata = el_strdup("{}"); - e->weight = 0.0; - e->confidence = 1.0; - int64_t now = engram_now_ms(); - e->created_at = now; e->updated_at = now; - e->layer_id = ENGRAM_LAYER_DEFAULT; - int64_t idx = g->edge_count; - g->edge_count++; - return idx; -} - -void dharma_strengthen(el_val_t cgi_id, el_val_t weight) { - const char* id = EL_CSTR(cgi_id); - if (!id || !*id) return; - char* base = NULL; char* url = NULL; - dharma_parse_id(id, &base, &url); - free(url); - int64_t ei = dharma_find_or_create_relation_edge(base, 1); - free(base); - if (ei < 0) return; - EngramStore* g = engram_get(); - double inc = engram_decode_score(weight); - if (!(inc >= 0.0)) inc = 0.0; - double w = g->edges[ei].weight + inc; - if (w < 0.0) w = 0.0; - if (w > 1.0) w = 1.0; - g->edges[ei].weight = w; - g->edges[ei].updated_at = engram_now_ms(); - g->edges[ei].last_fired = g->edges[ei].updated_at; -} - -el_val_t dharma_relationship(el_val_t cgi_id) { - const char* id = EL_CSTR(cgi_id); - if (!id || !*id) return el_from_float(0.0); - char* base = NULL; char* url = NULL; - dharma_parse_id(id, &base, &url); - free(url); - int64_t ei = dharma_find_or_create_relation_edge(base, 0); - free(base); - if (ei < 0) return el_from_float(0.0); - EngramStore* g = engram_get(); - return el_from_float(g->edges[ei].weight); -} - -el_val_t dharma_peers(void) { - /* Walk dharma-relation edges out of "dharma:self", weight > 0, sort desc. */ - EngramStore* g = engram_get(); - const char* self_id = "dharma:self"; - typedef struct { char* peer_base; double weight; } PeerEntry; - PeerEntry* peers = malloc((size_t)(g->edge_count + 1) * sizeof(PeerEntry)); - int64_t pcount = 0; - if (!peers) return el_list_empty(); - for (int64_t i = 0; i < g->edge_count; i++) { - EngramEdge* e = &g->edges[i]; - if (!e->from_id || !e->to_id) continue; - if (strcmp(e->from_id, self_id) != 0) continue; - if (!e->relation || strcmp(e->relation, "dharma-relation") != 0) continue; - if (e->weight <= 0.0) continue; - const char* prefix = "dharma:peer:"; - size_t pl = strlen(prefix); - if (strncmp(e->to_id, prefix, pl) != 0) continue; - peers[pcount].peer_base = el_strdup(e->to_id + pl); - peers[pcount].weight = e->weight; - pcount++; - } - /* Sort desc by weight. */ - for (int64_t i = 1; i < pcount; i++) { - PeerEntry key = peers[i]; - int64_t j = i - 1; - while (j >= 0 && peers[j].weight < key.weight) { - peers[j + 1] = peers[j]; j--; - } - peers[j + 1] = key; - } - el_val_t out = el_list_empty(); - for (int64_t i = 0; i < pcount; i++) { - out = el_list_append(out, EL_STR(peers[i].peer_base)); - } - free(peers); - return out; -} - -/* ── Batch 4: LLM (Anthropic API client) ─────────────────────────────────── */ -/* - * All LLM builtins call https://api.anthropic.com/v1/messages with the API - * key from env ANTHROPIC_API_KEY. Default model is "claude-sonnet-4-5" - * when the supplied model is empty/null. - * - * `llm_call_agentic` runs a real multi-turn tool_use/tool_result loop. - * Tool handlers are registered with `llm_register_tool(name, fn_name)`, - * which dlsym()s the named symbol. Each tool handler has the C signature - * el_val_t handler(el_val_t input_json); - * and returns a JSON-string el_val_t result. Iteration is capped at 10. - */ - -static const char* LLM_DEFAULT_MODEL = "claude-sonnet-4-5"; -static const char* LLM_API_URL = "https://api.anthropic.com/v1/messages"; -static const char* LLM_VERSION = "2023-06-01"; - -static const char* llm_resolve_model(const char* m) { - if (!m || !*m) return LLM_DEFAULT_MODEL; - return m; -} - -/* - * ── Configurable LLM provider chain ────────────────────────────────────────── - * - * Providers are configured via indexed env vars. The runtime tries each in - * order (0, 1, 2, ...) and returns the first successful non-empty response. - * - * Per provider (N = 0, 1, 2, ...): - * NEURON_LLM_N_URL — endpoint URL (base URL; /v1/chat/completions appended - * if format is "openai" and not already in URL) - * NEURON_LLM_N_KEY — API key - * NEURON_LLM_N_FORMAT — "openai" (default) or "anthropic" - * NEURON_LLM_N_MODEL — model name override (optional) - * - * Example — Neuron inference primary, Anthropic fallback: - * NEURON_LLM_0_URL=https://soma.../v1/chat/completions - * NEURON_LLM_0_KEY=svc-key - * NEURON_LLM_0_FORMAT=openai - * NEURON_LLM_0_MODEL=neuron - * NEURON_LLM_1_URL=https://api.anthropic.com/v1/messages - * NEURON_LLM_1_KEY=sk-ant-... - * NEURON_LLM_1_FORMAT=anthropic - * - * If no NEURON_LLM_0_URL is set, falls back to legacy ANTHROPIC_API_KEY. - */ - -#define LLM_MAX_PROVIDERS 16 - -/* forward declarations */ -static el_val_t llm_extract_text(el_val_t resp_val); -static el_val_t llm_extract_text_openai(el_val_t resp_val); - -static el_val_t llm_extract_text_openai(el_val_t resp_val) { - const char* resp = EL_CSTR(resp_val); - if (!resp || !*resp) return el_wrap_str(el_strdup("")); - if (resp[0] == '{' && strstr(resp, "\"error\"")) return el_wrap_str(el_strdup("")); - const char* choices = json_find_key(resp, "choices"); - if (!choices || *choices != '[') return el_wrap_str(el_strdup("")); - choices++; - while (*choices == ' ' || *choices == '\t') choices++; - if (*choices != '{') return el_wrap_str(el_strdup("")); - const char* end = json_skip_value(choices); - size_t n = (size_t)(end - choices); - char* obj = malloc(n + 1); memcpy(obj, choices, n); obj[n] = '\0'; - const char* msg = json_find_key(obj, "message"); - if (!msg || *msg != '{') { free(obj); return el_wrap_str(el_strdup("")); } - const char* msg_end = json_skip_value(msg); - size_t mn = (size_t)(msg_end - msg); - char* msg_obj = malloc(mn + 1); memcpy(msg_obj, msg, mn); msg_obj[mn] = '\0'; - const char* content = json_find_key(msg_obj, "content"); - el_val_t result = el_wrap_str(el_strdup("")); - if (content && *content == '"') { - JsonParser jp = { .p = content, .end = content + strlen(content), .err = 0 }; - char* text = jp_parse_string_raw(&jp); - if (!jp.err && text) result = el_wrap_str(text); - } - free(msg_obj); free(obj); - return result; -} - -/* Send a request to one provider. Returns the raw response string. - * format: 0 = openai, 1 = anthropic */ -static el_val_t llm_provider_request(const char* url, const char* key, - int format, const char* model, - const char* system_str, - const char* user_str) { - char* esc_sys = system_str && *system_str ? json_escape_alloc(system_str) : NULL; - char* esc_user = json_escape_alloc(user_str ? user_str : ""); - JsonBuf b; jb_init(&b); - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - - if (format == 0) { /* OpenAI */ - char full_url[1024]; - if (strstr(url, "/chat/completions") || strstr(url, "/messages")) { - snprintf(full_url, sizeof(full_url), "%s", url); - } else { - snprintf(full_url, sizeof(full_url), "%s/v1/chat/completions", url); - } - { size_t n = strlen(key)+24; char* l=malloc(n); snprintf(l,n,"Authorization: Bearer %s",key); h=curl_slist_append(h,l); free(l); } - jb_putc(&b, '{'); - jb_puts(&b, "\"model\":"); jb_emit_escaped(&b, model ? model : "neuron"); - jb_puts(&b, ",\"max_tokens\":4096,\"messages\":["); - if (esc_sys && *esc_sys) { jb_puts(&b,"{\"role\":\"system\",\"content\":\""); jb_puts(&b,esc_sys); jb_puts(&b,"\"},"); } - jb_puts(&b, "{\"role\":\"user\",\"content\":\""); jb_puts(&b, esc_user); jb_puts(&b, "\"}]}"); - el_val_t resp = http_do("POST", full_url, b.buf, h); - curl_slist_free_all(h); free(b.buf); - if (esc_sys) free(esc_sys); free(esc_user); - return llm_extract_text_openai(resp); - } else { /* Anthropic */ - { size_t n = strlen(key)+16; char* l=malloc(n); snprintf(l,n,"x-api-key: %s",key); h=curl_slist_append(h,l); free(l); } - { size_t n = strlen(LLM_VERSION)+32; char* l=malloc(n); snprintf(l,n,"anthropic-version: %s",LLM_VERSION); h=curl_slist_append(h,l); free(l); } - jb_putc(&b, '{'); - jb_puts(&b, "\"model\":"); jb_emit_escaped(&b, model ? model : LLM_DEFAULT_MODEL); - jb_puts(&b, ",\"max_tokens\":4096"); - if (esc_sys && *esc_sys) { jb_puts(&b,",\"system\":\""); jb_puts(&b,esc_sys); jb_puts(&b,"\""); } - jb_puts(&b, ",\"messages\":[{\"role\":\"user\",\"content\":\""); jb_puts(&b, esc_user); jb_puts(&b, "\"}]}"); - el_val_t resp = http_do("POST", url, b.buf, h); - curl_slist_free_all(h); free(b.buf); - if (esc_sys) free(esc_sys); free(esc_user); - return llm_extract_text(resp); - } -} - -static el_val_t llm_chain_call(const char* system_str, const char* user_str) { - char url_key[64], key_key[64], fmt_key[64], model_key[64]; - for (int i = 0; i < LLM_MAX_PROVIDERS; i++) { - snprintf(url_key, sizeof(url_key), "NEURON_LLM_%d_URL", i); - snprintf(key_key, sizeof(key_key), "NEURON_LLM_%d_KEY", i); - snprintf(fmt_key, sizeof(fmt_key), "NEURON_LLM_%d_FORMAT", i); - snprintf(model_key, sizeof(model_key), "NEURON_LLM_%d_MODEL", i); - const char* url = getenv(url_key); - const char* key = getenv(key_key); - if (!url || !*url || !key || !*key) break; /* end of chain */ - const char* fmt_s = getenv(fmt_key); - int fmt = (fmt_s && strcmp(fmt_s, "anthropic") == 0) ? 1 : 0; - const char* model = getenv(model_key); - fprintf(stderr, "[llm] trying provider %d (%s)\n", i, url); - el_val_t result = llm_provider_request(url, key, fmt, model, system_str, user_str); - const char* t = EL_CSTR(result); - if (t && *t && t[0] != '{') return result; /* success */ - fprintf(stderr, "[llm] provider %d failed or empty, trying next\n", i); - } - /* Legacy fallback: ANTHROPIC_API_KEY */ - const char* api_key = getenv("ANTHROPIC_API_KEY"); - if (!api_key || !*api_key) return http_error_json("no LLM providers configured"); - fprintf(stderr, "[llm] using legacy ANTHROPIC_API_KEY fallback\n"); - return llm_provider_request(LLM_API_URL, api_key, 1, NULL, system_str, user_str); -} - -/* Legacy llm_request — kept for backward compat with agentic loop internals */ -static el_val_t llm_request(const char* json_body) { - const char* api_key = getenv("ANTHROPIC_API_KEY"); - if (!api_key || !*api_key) return http_error_json("ANTHROPIC_API_KEY not set"); - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - { size_t n=strlen(api_key)+16; char* l=malloc(n); snprintf(l,n,"x-api-key: %s",api_key); h=curl_slist_append(h,l); free(l); } - { size_t n=strlen(LLM_VERSION)+32; char* l=malloc(n); snprintf(l,n,"anthropic-version: %s",LLM_VERSION); h=curl_slist_append(h,l); free(l); } - el_val_t resp = http_do("POST", LLM_API_URL, json_body, h); - curl_slist_free_all(h); - return resp; -} - -/* Extract concatenated assistant text from an Anthropic /v1/messages - * response. The response shape is: - * {"content":[{"type":"text","text":"..."}, ...], ...} - * If parsing fails, returns the raw response so the caller can inspect. - */ -static el_val_t llm_extract_text(el_val_t resp_val) { - const char* resp = EL_CSTR(resp_val); - if (!resp || !*resp) return el_wrap_str(el_strdup("")); - /* If error JSON, propagate as-is. */ - if (resp[0] == '{' && strstr(resp, "\"error\"")) { - return el_wrap_str(el_strdup(resp)); - } - /* Find "content":[ ... ] */ - const char* p = json_find_key(resp, "content"); - if (!p) return el_wrap_str(el_strdup(resp)); - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '[') return el_wrap_str(el_strdup(resp)); - p++; - JsonBuf out; jb_init(&out); - while (*p && *p != ']') { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - if (*p != '{') break; - const char* end = json_skip_value(p); - size_t n = (size_t)(end - p); - char* obj = malloc(n + 1); - memcpy(obj, p, n); obj[n] = '\0'; - const char* type_p = json_find_key(obj, "type"); - if (type_p && *type_p == '"') { - JsonParser jp = { .p = type_p, .end = type_p + strlen(type_p), .err = 0 }; - char* type_s = jp_parse_string_raw(&jp); - if (!jp.err && type_s && strcmp(type_s, "text") == 0) { - const char* tp = json_find_key(obj, "text"); - if (tp && *tp == '"') { - JsonParser jp2 = { .p = tp, .end = tp + strlen(tp), .err = 0 }; - char* text_s = jp_parse_string_raw(&jp2); - if (!jp2.err && text_s) jb_puts(&out, text_s); - free(text_s); - } - } - free(type_s); - } - free(obj); - p = end; - } - return el_wrap_str(out.buf); -} - -el_val_t llm_call(el_val_t model, el_val_t prompt) { - const char* u = EL_CSTR(prompt); if (!u) u = ""; - return llm_chain_call(NULL, u); -} - -el_val_t llm_call_system(el_val_t model, el_val_t system_prompt, el_val_t user_prompt) { - const char* s = EL_CSTR(system_prompt); if (!s) s = ""; - const char* u = EL_CSTR(user_prompt); if (!u) u = ""; - return llm_chain_call(s, u); -} - -/* ── Tool registry for llm_call_agentic ─────────────────────────────────── */ - -typedef el_val_t (*llm_tool_fn)(el_val_t input); - -typedef struct LlmToolEntry { - char* name; - llm_tool_fn fn; -} LlmToolEntry; - -static LlmToolEntry _llm_tools[64]; -static size_t _llm_tool_count = 0; -static pthread_mutex_t _llm_tool_mu = PTHREAD_MUTEX_INITIALIZER; - -static llm_tool_fn llm_tool_lookup(const char* name) { - if (!name) return NULL; - llm_tool_fn fn = NULL; - pthread_mutex_lock(&_llm_tool_mu); - for (size_t i = 0; i < _llm_tool_count; i++) { - if (strcmp(_llm_tools[i].name, name) == 0) { fn = _llm_tools[i].fn; break; } - } - pthread_mutex_unlock(&_llm_tool_mu); - return fn; -} - -void llm_register_tool(el_val_t name, el_val_t handler_fn_name) { - const char* nm = EL_CSTR(name); - const char* sym = EL_CSTR(handler_fn_name); - if (!nm || !*nm || !sym || !*sym) return; - void* p = dlsym(RTLD_DEFAULT, sym); - if (!p) { - fprintf(stderr, "[llm_register_tool] symbol not found: %s\n", sym); - return; - } - pthread_mutex_lock(&_llm_tool_mu); - /* Replace existing entry by name. */ - for (size_t i = 0; i < _llm_tool_count; i++) { - if (strcmp(_llm_tools[i].name, nm) == 0) { - _llm_tools[i].fn = (llm_tool_fn)p; - pthread_mutex_unlock(&_llm_tool_mu); - return; - } - } - if (_llm_tool_count < sizeof(_llm_tools) / sizeof(_llm_tools[0])) { - _llm_tools[_llm_tool_count].name = el_strdup(nm); - _llm_tools[_llm_tool_count].fn = (llm_tool_fn)p; - _llm_tool_count++; - } - pthread_mutex_unlock(&_llm_tool_mu); -} - -/* Serialize the El `tools` list into the JSON `tools:[...]` field expected - * by the Anthropic API. Each tool is an ElMap with name/description/ - * input_schema. input_schema is treated as either a JSON-object string - * (passed through verbatim) or a missing field (substitute {}). */ -static void llm_emit_tools_json(JsonBuf* b, el_val_t tools_list) { - jb_putc(b, '['); - ElList* lst = (ElList*)(uintptr_t)tools_list; - int64_t n = lst ? lst->length : 0; - for (int64_t i = 0; i < n; i++) { - if (i > 0) jb_putc(b, ','); - ElMap* tm = as_map(lst->elems[i]); - const char* name = ""; - const char* desc = ""; - const char* schema = "{}"; - if (tm) { - for (int64_t k = 0; k < tm->count; k++) { - const char* key = EL_CSTR(tm->keys[k]); - const char* val = EL_CSTR(tm->values[k]); - if (!key || !val) continue; - if (strcmp(key, "name") == 0) name = val; - else if (strcmp(key, "description") == 0) desc = val; - else if (strcmp(key, "input_schema") == 0) schema = val; - } - } - char* esc_name = json_escape_alloc(name); - char* esc_desc = json_escape_alloc(desc); - jb_puts(b, "{\"name\":\""); jb_puts(b, esc_name); - jb_puts(b, "\",\"description\":\""); jb_puts(b, esc_desc); - jb_puts(b, "\",\"input_schema\":"); jb_puts(b, schema && *schema ? schema : "{}"); - jb_putc(b, '}'); - free(esc_name); free(esc_desc); - } - jb_putc(b, ']'); -} - -/* Walk the assistant `content` array and emit each block back into b, - * preserving the verbatim JSON of every block — used to re-include the - * assistant turn in the next request. */ -static void llm_emit_content_blocks(JsonBuf* b, const char* resp) { - const char* p = json_find_key(resp, "content"); - jb_putc(b, '['); - if (!p) { jb_putc(b, ']'); return; } - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '[') { jb_putc(b, ']'); return; } - p++; - int first = 1; - while (*p && *p != ']') { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - if (*p != '{') break; - const char* end = json_skip_value(p); - if (!first) jb_putc(b, ','); - first = 0; - size_t n = (size_t)(end - p); - jb_reserve(b, n); - memcpy(b->buf + b->len, p, n); - b->len += n; - b->buf[b->len] = '\0'; - p = end; - } - jb_putc(b, ']'); -} - -/* Concatenate all "text" blocks from a response. Returns owned string. */ -static char* llm_concat_text_blocks(const char* resp) { - JsonBuf out; jb_init(&out); - if (!resp) return out.buf; - const char* p = json_find_key(resp, "content"); - if (!p) return out.buf; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '[') return out.buf; - p++; - while (*p && *p != ']') { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - if (*p != '{') break; - const char* end = json_skip_value(p); - size_t n = (size_t)(end - p); - char* obj = malloc(n + 1); - memcpy(obj, p, n); obj[n] = '\0'; - const char* tp = json_find_key(obj, "type"); - if (tp && *tp == '"') { - JsonParser jp = { .p = tp, .end = tp + strlen(tp), .err = 0 }; - char* tname = jp_parse_string_raw(&jp); - if (!jp.err && tname && strcmp(tname, "text") == 0) { - const char* xp = json_find_key(obj, "text"); - if (xp && *xp == '"') { - JsonParser jp2 = { .p = xp, .end = xp + strlen(xp), .err = 0 }; - char* txt = jp_parse_string_raw(&jp2); - if (!jp2.err && txt) jb_puts(&out, txt); - free(txt); - } - } - free(tname); - } - free(obj); - p = end; - } - return out.buf; -} - -/* Build tool_result message blocks for every tool_use in a response. - * Appends to `b` an array element for each tool_use; caller wraps. */ -static int llm_build_tool_results(JsonBuf* b, const char* resp) { - int any = 0; - const char* p = json_find_key(resp, "content"); - if (!p) return 0; - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; - if (*p != '[') return 0; - p++; - while (*p && *p != ']') { - while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r' || *p == ',') p++; - if (*p != '{') break; - const char* end = json_skip_value(p); - size_t n = (size_t)(end - p); - char* obj = malloc(n + 1); - memcpy(obj, p, n); obj[n] = '\0'; - - const char* tp = json_find_key(obj, "type"); - char* type_s = NULL; - if (tp && *tp == '"') { - JsonParser jp = { .p = tp, .end = tp + strlen(tp), .err = 0 }; - type_s = jp_parse_string_raw(&jp); - } - if (type_s && strcmp(type_s, "tool_use") == 0) { - /* Extract id, name, input. */ - char* id_s = NULL; char* name_s = NULL; - const char* idp = json_find_key(obj, "id"); - if (idp && *idp == '"') { - JsonParser jp = { .p = idp, .end = idp + strlen(idp), .err = 0 }; - id_s = jp_parse_string_raw(&jp); - } - const char* np = json_find_key(obj, "name"); - if (np && *np == '"') { - JsonParser jp = { .p = np, .end = np + strlen(np), .err = 0 }; - name_s = jp_parse_string_raw(&jp); - } - el_val_t input_raw = json_get_raw(EL_STR(obj), EL_STR("input")); - const char* input_s = EL_CSTR(input_raw); - if (!input_s || !*input_s) input_s = "{}"; - - llm_tool_fn fn = llm_tool_lookup(name_s ? name_s : ""); - char* result = NULL; - int is_error = 0; - if (!fn) { - size_t en = strlen(name_s ? name_s : "(null)") + 64; - result = malloc(en); - snprintf(result, en, "{\"error\":\"tool not registered: %s\"}", - name_s ? name_s : "(null)"); - is_error = 1; - } else { - el_val_t out = fn(EL_STR(input_s)); - const char* os = EL_CSTR(out); - result = el_strdup(os ? os : ""); - } - - if (any) jb_putc(b, ','); - char* esc_id = json_escape_alloc(id_s ? id_s : ""); - char* esc_res = json_escape_alloc(result ? result : ""); - jb_puts(b, "{\"type\":\"tool_result\",\"tool_use_id\":\""); - jb_puts(b, esc_id); - jb_puts(b, "\",\"content\":\""); - jb_puts(b, esc_res); - jb_puts(b, "\""); - if (is_error) jb_puts(b, ",\"is_error\":true"); - jb_putc(b, '}'); - free(esc_id); free(esc_res); free(result); - free(id_s); free(name_s); - any = 1; - } - free(type_s); - free(obj); - p = end; - } - return any; -} - -el_val_t llm_call_agentic(el_val_t model, el_val_t system, el_val_t user, el_val_t tools) { - /* Empty tools list → degrade to plain system call. */ - ElList* tl = (ElList*)(uintptr_t)tools; - if (!tl || tl->length == 0) { - return llm_call_system(model, system, user); - } - const char* m = llm_resolve_model(EL_CSTR(model)); - const char* sys_p = EL_CSTR(system); if (!sys_p) sys_p = ""; - const char* usr_p = EL_CSTR(user); if (!usr_p) usr_p = ""; - - /* Build the static parts: tools JSON and system prompt — these don't - * change across iterations. */ - JsonBuf tools_buf; jb_init(&tools_buf); - llm_emit_tools_json(&tools_buf, tools); - char* esc_sys = json_escape_alloc(sys_p); - - /* messages array, accumulated as a mutable JSON fragment (no surrounding - * brackets — emitted at request time). */ - JsonBuf msgs; jb_init(&msgs); - /* First user message. */ - char* esc_user = json_escape_alloc(usr_p); - jb_puts(&msgs, "{\"role\":\"user\",\"content\":\""); - jb_puts(&msgs, esc_user); - jb_puts(&msgs, "\"}"); - free(esc_user); - - char* last_text = el_strdup(""); - el_val_t final_out = 0; - int reached_cap = 1; - - for (int iter = 0; iter < 10; iter++) { - /* Build request body. */ - JsonBuf body; jb_init(&body); - jb_putc(&body, '{'); - jb_puts(&body, "\"model\":"); jb_emit_escaped(&body, m); - jb_puts(&body, ",\"max_tokens\":4096"); - if (*sys_p) { - jb_puts(&body, ",\"system\":\""); - jb_puts(&body, esc_sys); - jb_puts(&body, "\""); - } - jb_puts(&body, ",\"tools\":"); - jb_puts(&body, tools_buf.buf); - jb_puts(&body, ",\"messages\":["); - jb_puts(&body, msgs.buf); - jb_puts(&body, "]}"); - - el_val_t resp_v = llm_request(body.buf); - free(body.buf); - const char* resp = EL_CSTR(resp_v); - if (!resp || !*resp) { - final_out = http_error_json("empty response"); - reached_cap = 0; - break; - } - if (resp[0] == '{' && strstr(resp, "\"error\"") && - !json_find_key(resp, "content")) { - final_out = el_wrap_str(el_strdup(resp)); - reached_cap = 0; - break; - } - - /* Update last_text from this response. */ - free(last_text); - last_text = llm_concat_text_blocks(resp); - - /* Inspect stop_reason. */ - el_val_t sr_v = json_get_string(EL_STR(resp), EL_STR("stop_reason")); - const char* sr = EL_CSTR(sr_v); if (!sr) sr = ""; - - if (strcmp(sr, "end_turn") == 0) { - final_out = el_wrap_str(el_strdup(last_text)); - reached_cap = 0; - break; - } - if (strcmp(sr, "max_tokens") == 0) { - size_t ln = strlen(last_text) + 16; - char* out = malloc(ln); - snprintf(out, ln, "%s\n[truncated]", last_text); - final_out = el_wrap_str(out); - reached_cap = 0; - break; - } - if (strcmp(sr, "tool_use") != 0) { - /* Unexpected stop reason; return the text we have. */ - final_out = el_wrap_str(el_strdup(last_text)); - reached_cap = 0; - break; - } - - /* Append the assistant turn (raw content blocks) to messages. */ - JsonBuf ab; jb_init(&ab); - jb_puts(&ab, ",{\"role\":\"assistant\",\"content\":"); - llm_emit_content_blocks(&ab, resp); - jb_putc(&ab, '}'); - jb_puts(&msgs, ab.buf); - free(ab.buf); - - /* Build tool_result message. */ - JsonBuf tr; jb_init(&tr); - jb_puts(&tr, ",{\"role\":\"user\",\"content\":["); - int any = llm_build_tool_results(&tr, resp); - jb_puts(&tr, "]}"); - if (any) { - jb_puts(&msgs, tr.buf); - } - free(tr.buf); - } - - if (reached_cap) { - size_t ln = strlen(last_text) + 32; - char* out = malloc(ln); - snprintf(out, ln, "[loop_cap_reached]\n%s", last_text); - final_out = el_wrap_str(out); - } - free(last_text); - free(esc_sys); - free(tools_buf.buf); - free(msgs.buf); - return final_out; -} - -/* base64-encode arbitrary bytes (returns owned C string). - * Internal helper for llm_vision; the public crypto entry point that El - * programs call is `base64_encode(el_val_t)` defined in the crypto block - * at the end of this file. */ -static char* el_b64_encode_internal(const unsigned char* src, size_t n) { - static const char tbl[] = - "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; - size_t out_len = 4 * ((n + 2) / 3); - char* out = malloc(out_len + 1); - if (!out) return NULL; - size_t o = 0; - for (size_t i = 0; i < n;) { - uint32_t v = 0; int got = 0; - v |= (uint32_t)src[i++] << 16; got++; - if (i < n) { v |= (uint32_t)src[i++] << 8; got++; } - if (i < n) { v |= (uint32_t)src[i++]; got++; } - out[o++] = tbl[(v >> 18) & 0x3f]; - out[o++] = tbl[(v >> 12) & 0x3f]; - out[o++] = (got > 1) ? tbl[(v >> 6) & 0x3f] : '='; - out[o++] = (got > 2) ? tbl[v & 0x3f] : '='; - } - out[o] = '\0'; - return out; -} - -el_val_t llm_vision(el_val_t model, el_val_t system, el_val_t prompt, el_val_t image_url_or_b64) { - const char* m = llm_resolve_model(EL_CSTR(model)); - const char* s = EL_CSTR(system); if (!s) s = ""; - const char* u = EL_CSTR(prompt); if (!u) u = ""; - const char* img = EL_CSTR(image_url_or_b64); if (!img) img = ""; - - /* Choose source mode */ - char* image_block = NULL; - if (strncasecmp(img, "http://", 7) == 0 || strncasecmp(img, "https://", 8) == 0) { - char* esc_url = json_escape_alloc(img); - size_t n = strlen(esc_url) + 128; - image_block = malloc(n); - snprintf(image_block, n, - "{\"type\":\"image\",\"source\":{\"type\":\"url\",\"url\":\"%s\"}}", - esc_url); - free(esc_url); - } else if (strncmp(img, "data:", 5) == 0) { - /* Inline data URL: split media-type and base64 */ - const char* semi = strchr(img + 5, ';'); - const char* comma = strchr(img + 5, ','); - char media[64] = "image/png"; - if (semi && comma && semi < comma) { - size_t ml = (size_t)(semi - (img + 5)); - if (ml >= sizeof(media)) ml = sizeof(media) - 1; - memcpy(media, img + 5, ml); media[ml] = '\0'; - } - const char* b64 = comma ? comma + 1 : ""; - char* esc_media = json_escape_alloc(media); - char* esc_b64 = json_escape_alloc(b64); - size_t n = strlen(esc_media) + strlen(esc_b64) + 192; - image_block = malloc(n); - snprintf(image_block, n, - "{\"type\":\"image\",\"source\":{\"type\":\"base64\"," - "\"media_type\":\"%s\",\"data\":\"%s\"}}", - esc_media, esc_b64); - free(esc_media); free(esc_b64); - } else if (*img) { - /* Treat as file path: read, base64-encode, attach. */ - FILE* f = fopen(img, "rb"); - if (!f) { - char err[256]; snprintf(err, sizeof(err), "cannot open image: %s", img); - return http_error_json(err); - } - fseek(f, 0, SEEK_END); - long sz = ftell(f); - rewind(f); - if (sz <= 0) { fclose(f); return http_error_json("empty image file"); } - unsigned char* buf = malloc((size_t)sz); - if (!buf) { fclose(f); return http_error_json("oom"); } - size_t got = fread(buf, 1, (size_t)sz, f); - fclose(f); - char* b64 = el_b64_encode_internal(buf, got); - free(buf); - if (!b64) return http_error_json("base64 encode failed"); - const char* media = "image/png"; - size_t ilen = strlen(img); - if (ilen >= 4) { - if (strcasecmp(img + ilen - 4, ".jpg") == 0 || - (ilen >= 5 && strcasecmp(img + ilen - 5, ".jpeg") == 0)) media = "image/jpeg"; - else if (strcasecmp(img + ilen - 4, ".gif") == 0) media = "image/gif"; - else if (strcasecmp(img + ilen - 4, ".webp") == 0) media = "image/webp"; - } - char* esc_b64 = json_escape_alloc(b64); free(b64); - size_t n = strlen(esc_b64) + 192; - image_block = malloc(n); - snprintf(image_block, n, - "{\"type\":\"image\",\"source\":{\"type\":\"base64\"," - "\"media_type\":\"%s\",\"data\":\"%s\"}}", - media, esc_b64); - free(esc_b64); - } - - char* esc_sys = json_escape_alloc(s); - char* esc_user = json_escape_alloc(u); - JsonBuf b; jb_init(&b); - jb_putc(&b, '{'); - jb_puts(&b, "\"model\":"); jb_emit_escaped(&b, m); - jb_puts(&b, ",\"max_tokens\":4096"); - if (*s) { - jb_puts(&b, ",\"system\":\""); - jb_puts(&b, esc_sys); - jb_puts(&b, "\""); - } - jb_puts(&b, ",\"messages\":[{\"role\":\"user\",\"content\":["); - if (image_block) { - jb_puts(&b, image_block); - jb_putc(&b, ','); - } - jb_puts(&b, "{\"type\":\"text\",\"text\":\""); - jb_puts(&b, esc_user); - jb_puts(&b, "\"}]}]}"); - free(esc_sys); free(esc_user); free(image_block); - el_val_t resp = llm_request(b.buf); - free(b.buf); - return llm_extract_text(resp); -} - -el_val_t llm_models(void) { - el_val_t lst = el_list_empty(); - lst = el_list_append(lst, el_wrap_str(el_strdup("claude-sonnet-4-5"))); - lst = el_list_append(lst, el_wrap_str(el_strdup("claude-opus-4-7"))); - lst = el_list_append(lst, el_wrap_str(el_strdup("claude-haiku-4-5"))); - return lst; -} - -/* ── Native VM builtin aliases ────────────────────────────────────────────── - * El source files use native_* names (El VM builtins). - * When compiled to C, these map directly to el_* runtime functions. */ - -el_val_t native_list_get(el_val_t list, el_val_t index) { - return el_list_get(list, index); -} - -el_val_t native_list_len(el_val_t list) { - return el_list_len(list); -} - -el_val_t native_list_append(el_val_t list, el_val_t elem) { - return el_list_append(list, elem); -} - -el_val_t native_list_empty(void) { - return el_list_empty(); -} - -el_val_t native_list_clone(el_val_t list) { - return el_list_clone(list); -} - -el_val_t native_string_chars(el_val_t sv) { - const char* s = EL_CSTR(sv); - el_val_t result = el_list_empty(); - if (!s) return result; - while (*s) { - char buf[2]; - buf[0] = *s; - buf[1] = '\0'; - result = el_list_append(result, EL_STR(strdup(buf))); - s++; - } - return result; -} - -el_val_t native_int_to_str(el_val_t n) { - return int_to_str(n); -} - -/* ── Method-call shorthand aliases ────────────────────────────────────────── - * Short names that result from the method-call convention: - * myList.append(x) → append(myList, x) - * myList.len() → len(myList) - * myList.get(i) → get(myList, i) - * myMap.map_get(k) → map_get(myMap, k) - * myMap.map_set(k,v) → map_set(myMap, k, v) */ - -el_val_t append(el_val_t list, el_val_t elem) { return el_list_append(list, elem); } -el_val_t len(el_val_t list) { return el_list_len(list); } -el_val_t get(el_val_t list, el_val_t index) { return el_list_get(list, index); } -el_val_t map_get(el_val_t map, el_val_t key) { return el_map_get(map, key); } -el_val_t map_set(el_val_t map, el_val_t key, el_val_t value) { return el_map_set(map, key, value); } - -/* ── Crypto primitives ────────────────────────────────────────────────────── - * - * SHA-256 implementation adapted from Brad Conte's public-domain reference - * (https://github.com/B-Con/crypto-algorithms/blob/master/sha256.c, public - * domain per the project's LICENSE). HMAC follows RFC 2104. Base64 encoding - * follows RFC 4648; the URL-safe variant uses the alphabet from §5 of the - * RFC and omits padding (per JWT/JWS convention). - * - * Self-contained: no OpenSSL/libcrypto dependency. The runtime keeps its - * existing `-lcurl -lpthread -ldl -lm` link line. - * - * Binary outputs (sha256_bytes, hmac_sha256_bytes) tag their buffer with a - * magic header so base64_encode/base64url_encode can recover the exact byte - * length even when the payload contains embedded NULs. Plain C strings - * (without the header) fall back to strlen(), preserving the existing API - * shape for normal text inputs. */ - -/* Magic-header for length-tagged binary buffers. Layout: - * [ uint32_t magic = EL_MAGIC_BIN ][ uint32_t length ][ data... ][ \0 ] - * The returned el_val_t points at `data`, so consumers that strlen() it still - * get a sensible (though possibly truncated) view. el_bin_len() recovers the - * true length by sniffing the 8 bytes preceding the pointer. - * - * Magic value chosen with high MSB so it cannot collide with printable ASCII - * (the same discriminator pattern used by EL_MAGIC_LIST / EL_MAGIC_MAP). */ -#define EL_MAGIC_BIN 0xE1B17EAFu - -typedef struct { - uint32_t magic; - uint32_t length; -} el_bin_hdr_t; - -/* Allocate a length-tagged binary buffer; returns pointer to the data area. */ -static unsigned char* el_bin_alloc(size_t len) { - el_bin_hdr_t* hdr = (el_bin_hdr_t*)malloc(sizeof(el_bin_hdr_t) + len + 1); - if (!hdr) { fputs("el_runtime: out of memory (bin)\n", stderr); exit(1); } - hdr->magic = EL_MAGIC_BIN; - hdr->length = (uint32_t)len; - unsigned char* data = (unsigned char*)(hdr + 1); - data[len] = '\0'; /* keep NUL-terminated for accidental strlen calls */ - return data; -} - -/* Recover length from a possibly-tagged buffer. Returns 1 if tagged. */ -static int el_bin_lookup(const void* p, size_t* out_len) { - if (!p) { *out_len = 0; return 0; } - /* Avoid reading off the front of a page on tiny pointers (e.g. NULs - * passed in as int-cast values). 4096 is a safe lower bound on any - * platform we target. */ - if ((uintptr_t)p < 4096) return 0; - const el_bin_hdr_t* hdr = (const el_bin_hdr_t*)((const char*)p - sizeof(el_bin_hdr_t)); - if (hdr->magic != EL_MAGIC_BIN) return 0; - *out_len = hdr->length; - return 1; -} - -/* Effective input length: tagged length if present, else strlen. */ -static size_t el_input_len(const char* s) { - size_t n; - if (el_bin_lookup(s, &n)) return n; - return s ? strlen(s) : 0; -} - -/* ─── SHA-256 (Brad Conte / public domain) ──────────────────────────────── */ - -typedef struct { - unsigned char data[64]; - uint32_t datalen; - uint64_t bitlen; - uint32_t state[8]; -} el_sha256_ctx_t; - -static const uint32_t el_sha256_k[64] = { - 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5,0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5, - 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3,0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174, - 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc,0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da, - 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7,0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967, - 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13,0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85, - 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3,0xd192e819,0xd6990624,0xf40e3585,0x106aa070, - 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5,0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3, - 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208,0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2 -}; - -#define EL_ROTR(x, n) (((x) >> (n)) | ((x) << (32 - (n)))) -#define EL_CH(x,y,z) (((x) & (y)) ^ (~(x) & (z))) -#define EL_MAJ(x,y,z) (((x) & (y)) ^ ((x) & (z)) ^ ((y) & (z))) -#define EL_EP0(x) (EL_ROTR(x,2) ^ EL_ROTR(x,13) ^ EL_ROTR(x,22)) -#define EL_EP1(x) (EL_ROTR(x,6) ^ EL_ROTR(x,11) ^ EL_ROTR(x,25)) -#define EL_SIG0(x) (EL_ROTR(x,7) ^ EL_ROTR(x,18) ^ ((x) >> 3)) -#define EL_SIG1(x) (EL_ROTR(x,17) ^ EL_ROTR(x,19) ^ ((x) >> 10)) - -static void el_sha256_transform(el_sha256_ctx_t* ctx, const unsigned char* data) { - uint32_t a, b, c, d, e, f, g, h, t1, t2, m[64]; - int i, j; - for (i = 0, j = 0; i < 16; ++i, j += 4) { - m[i] = ((uint32_t)data[j] << 24) | ((uint32_t)data[j + 1] << 16) - | ((uint32_t)data[j + 2] << 8) | (uint32_t)data[j + 3]; - } - for (; i < 64; ++i) { - m[i] = EL_SIG1(m[i-2]) + m[i-7] + EL_SIG0(m[i-15]) + m[i-16]; - } - a = ctx->state[0]; b = ctx->state[1]; c = ctx->state[2]; d = ctx->state[3]; - e = ctx->state[4]; f = ctx->state[5]; g = ctx->state[6]; h = ctx->state[7]; - for (i = 0; i < 64; ++i) { - t1 = h + EL_EP1(e) + EL_CH(e,f,g) + el_sha256_k[i] + m[i]; - t2 = EL_EP0(a) + EL_MAJ(a,b,c); - h = g; g = f; f = e; e = d + t1; d = c; c = b; b = a; a = t1 + t2; - } - ctx->state[0] += a; ctx->state[1] += b; ctx->state[2] += c; ctx->state[3] += d; - ctx->state[4] += e; ctx->state[5] += f; ctx->state[6] += g; ctx->state[7] += h; -} - -static void el_sha256_init(el_sha256_ctx_t* ctx) { - ctx->datalen = 0; - ctx->bitlen = 0; - ctx->state[0] = 0x6a09e667; ctx->state[1] = 0xbb67ae85; - ctx->state[2] = 0x3c6ef372; ctx->state[3] = 0xa54ff53a; - ctx->state[4] = 0x510e527f; ctx->state[5] = 0x9b05688c; - ctx->state[6] = 0x1f83d9ab; ctx->state[7] = 0x5be0cd19; -} - -static void el_sha256_update(el_sha256_ctx_t* ctx, const unsigned char* data, size_t len) { - for (size_t i = 0; i < len; ++i) { - ctx->data[ctx->datalen++] = data[i]; - if (ctx->datalen == 64) { - el_sha256_transform(ctx, ctx->data); - ctx->bitlen += 512; - ctx->datalen = 0; - } - } -} - -static void el_sha256_final(el_sha256_ctx_t* ctx, unsigned char hash[32]) { - uint32_t i = ctx->datalen; - if (ctx->datalen < 56) { - ctx->data[i++] = 0x80; - while (i < 56) ctx->data[i++] = 0x00; - } else { - ctx->data[i++] = 0x80; - while (i < 64) ctx->data[i++] = 0x00; - el_sha256_transform(ctx, ctx->data); - memset(ctx->data, 0, 56); - } - ctx->bitlen += (uint64_t)ctx->datalen * 8; - ctx->data[63] = (unsigned char)( ctx->bitlen & 0xff); - ctx->data[62] = (unsigned char)((ctx->bitlen >> 8) & 0xff); - ctx->data[61] = (unsigned char)((ctx->bitlen >> 16) & 0xff); - ctx->data[60] = (unsigned char)((ctx->bitlen >> 24) & 0xff); - ctx->data[59] = (unsigned char)((ctx->bitlen >> 32) & 0xff); - ctx->data[58] = (unsigned char)((ctx->bitlen >> 40) & 0xff); - ctx->data[57] = (unsigned char)((ctx->bitlen >> 48) & 0xff); - ctx->data[56] = (unsigned char)((ctx->bitlen >> 56) & 0xff); - el_sha256_transform(ctx, ctx->data); - for (i = 0; i < 4; ++i) { - hash[i] = (ctx->state[0] >> (24 - i * 8)) & 0xff; - hash[i + 4] = (ctx->state[1] >> (24 - i * 8)) & 0xff; - hash[i + 8] = (ctx->state[2] >> (24 - i * 8)) & 0xff; - hash[i + 12] = (ctx->state[3] >> (24 - i * 8)) & 0xff; - hash[i + 16] = (ctx->state[4] >> (24 - i * 8)) & 0xff; - hash[i + 20] = (ctx->state[5] >> (24 - i * 8)) & 0xff; - hash[i + 24] = (ctx->state[6] >> (24 - i * 8)) & 0xff; - hash[i + 28] = (ctx->state[7] >> (24 - i * 8)) & 0xff; - } -} - -static void el_sha256_oneshot(const unsigned char* data, size_t len, unsigned char out[32]) { - el_sha256_ctx_t c; - el_sha256_init(&c); - el_sha256_update(&c, data, len); - el_sha256_final(&c, out); -} - -/* ─── HMAC-SHA-256 (RFC 2104) ───────────────────────────────────────────── */ - -static void el_hmac_sha256(const unsigned char* key, size_t key_len, - const unsigned char* msg, size_t msg_len, - unsigned char out[32]) { - unsigned char k[64]; - unsigned char k_ipad[64]; - unsigned char k_opad[64]; - unsigned char inner[32]; - - if (key_len > 64) { - el_sha256_oneshot(key, key_len, k); - memset(k + 32, 0, 32); - } else { - memcpy(k, key, key_len); - memset(k + key_len, 0, 64 - key_len); - } - for (int i = 0; i < 64; ++i) { - k_ipad[i] = k[i] ^ 0x36; - k_opad[i] = k[i] ^ 0x5c; - } - { - el_sha256_ctx_t c; - el_sha256_init(&c); - el_sha256_update(&c, k_ipad, 64); - el_sha256_update(&c, msg, msg_len); - el_sha256_final(&c, inner); - } - { - el_sha256_ctx_t c; - el_sha256_init(&c); - el_sha256_update(&c, k_opad, 64); - el_sha256_update(&c, inner, 32); - el_sha256_final(&c, out); - } -} - -/* ─── Hex helper ────────────────────────────────────────────────────────── */ - -static el_val_t el_hex_encode(const unsigned char* data, size_t len) { - static const char digits[] = "0123456789abcdef"; - char* out = el_strbuf(len * 2); - for (size_t i = 0; i < len; ++i) { - out[i * 2] = digits[(data[i] >> 4) & 0xf]; - out[i * 2 + 1] = digits[ data[i] & 0xf]; - } - out[len * 2] = '\0'; - return el_wrap_str(out); -} - -/* ─── Base64 (RFC 4648) ─────────────────────────────────────────────────── */ - -static const char el_b64_std_alphabet[64] = - "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; -static const char el_b64_url_alphabet[64] = - "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_"; - -el_val_t el_base64_encode_n(const unsigned char* data, size_t len, int url_safe) { - const char* alphabet = url_safe ? el_b64_url_alphabet : el_b64_std_alphabet; - /* Standard form is padded to multiple of 4; URL-safe omits padding. */ - size_t out_cap = ((len + 2) / 3) * 4 + 1; - char* out = el_strbuf(out_cap); - size_t i = 0, j = 0; - while (i + 3 <= len) { - uint32_t v = ((uint32_t)data[i] << 16) | ((uint32_t)data[i+1] << 8) | (uint32_t)data[i+2]; - out[j++] = alphabet[(v >> 18) & 0x3f]; - out[j++] = alphabet[(v >> 12) & 0x3f]; - out[j++] = alphabet[(v >> 6) & 0x3f]; - out[j++] = alphabet[ v & 0x3f]; - i += 3; - } - size_t rem = len - i; - if (rem == 1) { - uint32_t v = (uint32_t)data[i] << 16; - out[j++] = alphabet[(v >> 18) & 0x3f]; - out[j++] = alphabet[(v >> 12) & 0x3f]; - if (!url_safe) { out[j++] = '='; out[j++] = '='; } - } else if (rem == 2) { - uint32_t v = ((uint32_t)data[i] << 16) | ((uint32_t)data[i+1] << 8); - out[j++] = alphabet[(v >> 18) & 0x3f]; - out[j++] = alphabet[(v >> 12) & 0x3f]; - out[j++] = alphabet[(v >> 6) & 0x3f]; - if (!url_safe) { out[j++] = '='; } - } - out[j] = '\0'; - return el_wrap_str(out); -} - -/* Decode either alphabet — accepts both '+/' and '-_' transparently, and - * tolerates missing padding (which JWTs typically omit). Whitespace is - * skipped for robustness. Invalid characters cause the decode to stop and - * the partial result so far is returned. */ -static el_val_t el_base64_decode_any(const char* in) { - if (!in) { - unsigned char* empty = el_bin_alloc(0); - return EL_STR((char*)empty); - } - size_t in_len = strlen(in); - /* Worst case: 3 output bytes per 4 input chars, +1 NUL slack. */ - unsigned char* out = el_bin_alloc(((in_len + 3) / 4) * 3 + 1); - - int8_t lut[256]; - for (int i = 0; i < 256; ++i) lut[i] = -1; - for (int i = 0; i < 64; ++i) lut[(unsigned char)el_b64_std_alphabet[i]] = (int8_t)i; - /* Allow URL-safe characters too (so one decoder handles both forms). */ - lut[(unsigned char)'-'] = 62; - lut[(unsigned char)'_'] = 63; - - uint32_t buf = 0; - int bits = 0; - size_t o = 0; - for (size_t i = 0; i < in_len; ++i) { - unsigned char c = (unsigned char)in[i]; - if (c == '=' || c == '\r' || c == '\n' || c == ' ' || c == '\t') continue; - int8_t v = lut[c]; - if (v < 0) break; /* invalid char — stop */ - buf = (buf << 6) | (uint32_t)v; - bits += 6; - if (bits >= 8) { - bits -= 8; - out[o++] = (unsigned char)((buf >> bits) & 0xff); - } - } - /* Patch the length header to the actual decoded length. */ - el_bin_hdr_t* hdr = (el_bin_hdr_t*)((char*)out - sizeof(el_bin_hdr_t)); - hdr->length = (uint32_t)o; - out[o] = '\0'; - return EL_STR((char*)out); -} - -/* ─── Public crypto entry points ────────────────────────────────────────── */ - -el_val_t el_sha256_bytes_n(const unsigned char* data, size_t len) { - unsigned char* out = el_bin_alloc(32); - el_sha256_oneshot(data, len, out); - return EL_STR((char*)out); -} - -el_val_t sha256_hex(el_val_t input) { - const char* s = EL_CSTR(input); - size_t n = el_input_len(s); - unsigned char digest[32]; - el_sha256_oneshot((const unsigned char*)(s ? s : ""), n, digest); - return el_hex_encode(digest, 32); -} - -el_val_t sha256_bytes(el_val_t input) { - const char* s = EL_CSTR(input); - size_t n = el_input_len(s); - return el_sha256_bytes_n((const unsigned char*)(s ? s : ""), n); -} - -el_val_t hmac_sha256_hex(el_val_t key, el_val_t message) { - const char* k = EL_CSTR(key); - const char* m = EL_CSTR(message); - size_t kn = el_input_len(k); - size_t mn = el_input_len(m); - unsigned char mac[32]; - el_hmac_sha256((const unsigned char*)(k ? k : ""), kn, - (const unsigned char*)(m ? m : ""), mn, - mac); - return el_hex_encode(mac, 32); -} - -el_val_t hmac_sha256_bytes(el_val_t key, el_val_t message) { - const char* k = EL_CSTR(key); - const char* m = EL_CSTR(message); - size_t kn = el_input_len(k); - size_t mn = el_input_len(m); - unsigned char* out = el_bin_alloc(32); - el_hmac_sha256((const unsigned char*)(k ? k : ""), kn, - (const unsigned char*)(m ? m : ""), mn, - out); - return EL_STR((char*)out); -} - -el_val_t base64_encode(el_val_t input) { - const char* s = EL_CSTR(input); - size_t n = el_input_len(s); - return el_base64_encode_n((const unsigned char*)(s ? s : ""), n, /*url_safe=*/0); -} - -el_val_t base64url_encode(el_val_t input) { - const char* s = EL_CSTR(input); - size_t n = el_input_len(s); - return el_base64_encode_n((const unsigned char*)(s ? s : ""), n, /*url_safe=*/1); -} - -el_val_t base64_decode(el_val_t input) { - return el_base64_decode_any(EL_CSTR(input)); -} - -el_val_t base64url_decode(el_val_t input) { - return el_base64_decode_any(EL_CSTR(input)); -} - -/* ── Post-quantum cryptography (liboqs + OpenSSL) ─────────────────────────── - * - * Algorithm choices (per CNSA 2.0 / NIST PQ guidance, as of 2024): - * Signatures: CRYSTALS-Dilithium-3 (NIST security level 3, balanced) - * KEM: CRYSTALS-Kyber-768 (NIST security level 3) - * Hash: SHA3-256 (Keccak) (PQ-aware protocols favour SHA3 over SHA2) - * Hybrid: X25519 || Kyber-768, combined via HKDF-SHA256 - * - * Why hybrid: Kyber is new. X25519 has 20+ years of analysis. Hybridizing - * preserves classical security if Kyber falls to a future cryptanalytic - * advance, and preserves PQ security if X25519 falls to a quantum adversary. - * "Recordable now, decryptable later" already threatens long-lived classical - * key exchange — the only safe move for keys protecting durable doctrine - * (CGI lineage, KindredGrants, Principal-CGI covenants) is to encapsulate - * with PQ today, even if the classical leg is what the wire shows. - * - * Compile-time detection: when is unavailable the pq_* functions - * compile to stubs that return a JSON error envelope. SHA3-256 stays - * available regardless (it's implemented inline, no liboqs dep). This lets - * the runtime build cleanly on dev machines without liboqs while production - * gets the full PQ stack. */ - -/* ─── SHA3-256 (Keccak, FIPS 202) ──────────────────────────────────────────── - * Inline reference implementation. ~120 LoC, no external dependency. - * rate=1088 bits, capacity=512 bits, output=256 bits, padding=0x06. */ - -static const uint64_t el_keccak_rc[24] = { - 0x0000000000000001ULL, 0x0000000000008082ULL, 0x800000000000808aULL, - 0x8000000080008000ULL, 0x000000000000808bULL, 0x0000000080000001ULL, - 0x8000000080008081ULL, 0x8000000000008009ULL, 0x000000000000008aULL, - 0x0000000000000088ULL, 0x0000000080008009ULL, 0x000000008000000aULL, - 0x000000008000808bULL, 0x800000000000008bULL, 0x8000000000008089ULL, - 0x8000000000008003ULL, 0x8000000000008002ULL, 0x8000000000000080ULL, - 0x000000000000800aULL, 0x800000008000000aULL, 0x8000000080008081ULL, - 0x8000000000008080ULL, 0x0000000080000001ULL, 0x8000000080008008ULL -}; - -static const unsigned el_keccak_rho[24] = { - 1, 3, 6, 10, 15, 21, 28, 36, 45, 55, 2, 14, - 27, 41, 56, 8, 25, 43, 62, 18, 39, 61, 20, 44 -}; - -static const unsigned el_keccak_pi[24] = { - 10, 7, 11, 17, 18, 3, 5, 16, 8, 21, 24, 4, - 15, 23, 19, 13, 12, 2, 20, 14, 22, 9, 6, 1 -}; - -#define EL_ROTL64(x, n) (((x) << (n)) | ((x) >> (64 - (n)))) - -static void el_keccak_f1600(uint64_t s[25]) { - for (int round = 0; round < 24; ++round) { - uint64_t bc[5], t; - for (int i = 0; i < 5; ++i) - bc[i] = s[i] ^ s[i+5] ^ s[i+10] ^ s[i+15] ^ s[i+20]; - for (int i = 0; i < 5; ++i) { - t = bc[(i+4) % 5] ^ EL_ROTL64(bc[(i+1) % 5], 1); - for (int j = 0; j < 25; j += 5) s[j+i] ^= t; - } - t = s[1]; - for (int i = 0; i < 24; ++i) { - int j = el_keccak_pi[i]; - bc[0] = s[j]; - s[j] = EL_ROTL64(t, el_keccak_rho[i]); - t = bc[0]; - } - for (int j = 0; j < 25; j += 5) { - for (int i = 0; i < 5; ++i) bc[i] = s[j+i]; - for (int i = 0; i < 5; ++i) - s[j+i] = bc[i] ^ ((~bc[(i+1) % 5]) & bc[(i+2) % 5]); - } - s[0] ^= el_keccak_rc[round]; - } -} - -static void el_sha3_256_oneshot(const unsigned char* data, size_t len, - unsigned char out[32]) { - uint64_t st[25] = {0}; - unsigned char* sb = (unsigned char*)st; - const size_t rate = 136; /* 1088 bits / 8 */ - size_t i = 0; - while (len - i >= rate) { - for (size_t k = 0; k < rate; ++k) sb[k] ^= data[i + k]; - el_keccak_f1600(st); - i += rate; - } - size_t rem = len - i; - for (size_t k = 0; k < rem; ++k) sb[k] ^= data[i + k]; - sb[rem] ^= 0x06; /* SHA3 domain-separation byte */ - sb[rate - 1] ^= 0x80; /* final-block padding bit (high bit of last byte) */ - el_keccak_f1600(st); - memcpy(out, sb, 32); -} - -el_val_t sha3_256_hex(el_val_t input) { - const char* s = EL_CSTR(input); - size_t n = el_input_len(s); - unsigned char digest[32]; - el_sha3_256_oneshot((const unsigned char*)(s ? s : ""), n, digest); - return el_hex_encode(digest, 32); -} - -/* ─── Hex decode helper ───────────────────────────────────────────────────── - * Returns a length-tagged binary buffer (so embedded NULs survive); on - * odd-length / invalid input returns NULL with *out_len = 0. Caller is - * responsible for emitting the error envelope. */ - -static int el_hex_nibble(char c) { - if (c >= '0' && c <= '9') return c - '0'; - if (c >= 'a' && c <= 'f') return c - 'a' + 10; - if (c >= 'A' && c <= 'F') return c - 'A' + 10; - return -1; -} - -__attribute__((unused)) -static unsigned char* el_hex_decode(const char* s, size_t* out_len) { - *out_len = 0; - if (!s) return NULL; - size_t n = strlen(s); - if (n & 1) return NULL; - size_t blen = n / 2; - unsigned char* out = el_bin_alloc(blen); - for (size_t i = 0; i < blen; ++i) { - int hi = el_hex_nibble(s[i*2]); - int lo = el_hex_nibble(s[i*2 + 1]); - if (hi < 0 || lo < 0) return NULL; - out[i] = (unsigned char)((hi << 4) | lo); - } - *out_len = blen; - return out; -} - -/* JSON error envelope reused across all PQ entry points. */ -static el_val_t pq_error(const char* msg) { - return http_error_json(msg); -} - -#if __has_include() -#include -#define EL_HAVE_LIBOQS 1 -#else -#define EL_HAVE_LIBOQS 0 -#endif - -#if EL_HAVE_LIBOQS && __has_include() -#include -#define EL_HAVE_OPENSSL 1 -#else -#define EL_HAVE_OPENSSL 0 -#endif - -#if !EL_HAVE_LIBOQS - -/* ─── Stubs (liboqs unavailable) ─────────────────────────────────────────── - * Each entry point returns the same JSON error so callers can inspect a - * single canonical "missing primitive" string. pq_verify is the lone - * exception — verifying without liboqs simply means "not verified", so - * returning Bool false (0) keeps the type contract intact. */ - -#define EL_PQ_NO_LIB "liboqs not linked, post-quantum primitives unavailable" - -el_val_t pq_keygen_signature(void) { return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_sign(el_val_t sk, el_val_t msg) { (void)sk; (void)msg; return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_verify(el_val_t pk, el_val_t msg, el_val_t sig) { (void)pk; (void)msg; (void)sig; return EL_INT(0); } -el_val_t pq_kem_keygen(void) { return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_kem_encaps(el_val_t pk) { (void)pk; return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_kem_decaps(el_val_t sk, el_val_t ct) { (void)sk; (void)ct; return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_hybrid_keygen(void) { return pq_error(EL_PQ_NO_LIB); } -el_val_t pq_hybrid_handshake(el_val_t pub) { (void)pub; return pq_error(EL_PQ_NO_LIB); } - -#else /* EL_HAVE_LIBOQS */ - -/* ─── Dilithium-3 / ML-DSA-65 signatures ──────────────────────────────── - * - * NIST FIPS 204 standardized CRYSTALS-Dilithium as ML-DSA. ML-DSA-65 is the - * FIPS form of what we historically called Dilithium-3 — same algorithm - * family, same security level, identical key/sig sizes, but with a couple - * of standardization-driven tweaks (e.g. domain separation in the message - * binding). liboqs 0.12+ exposes both names; 0.15+ retired the legacy - * "Dilithium" constants in favour of "ML-DSA". We prefer ML-DSA-65 if the - * header advertises it, fall back to Dilithium-3 otherwise. Anything - * already signed with the older constant remains verifiable against that - * same constant — callers should pin the algorithm via the OQS_SIG handle's - * method_name field if they need to interoperate with archival signatures. */ - -#if defined(OQS_SIG_alg_ml_dsa_65) -# define EL_DILITHIUM_ALG OQS_SIG_alg_ml_dsa_65 -#elif defined(OQS_SIG_alg_dilithium_3) -# define EL_DILITHIUM_ALG OQS_SIG_alg_dilithium_3 -#else -# define EL_DILITHIUM_ALG "ML-DSA-65" /* string fallback; runtime probe catches misconfig */ -#endif - -el_val_t pq_keygen_signature(void) { - OQS_SIG* sig = OQS_SIG_new(EL_DILITHIUM_ALG); - if (!sig) return pq_error("OQS_SIG_new(dilithium-3) failed"); - unsigned char* pk = (unsigned char*)malloc(sig->length_public_key); - unsigned char* sk = (unsigned char*)malloc(sig->length_secret_key); - if (!pk || !sk) { free(pk); free(sk); OQS_SIG_free(sig); return pq_error("oom"); } - if (OQS_SIG_keypair(sig, pk, sk) != OQS_SUCCESS) { - free(pk); free(sk); OQS_SIG_free(sig); - return pq_error("dilithium-3 keypair generation failed"); - } - el_val_t pk_hex = el_hex_encode(pk, sig->length_public_key); - el_val_t sk_hex = el_hex_encode(sk, sig->length_secret_key); - OQS_MEM_secure_free(sk, sig->length_secret_key); - free(pk); - - const char* pks = EL_CSTR(pk_hex); - const char* sks = EL_CSTR(sk_hex); - char* buf = el_strbuf(strlen(pks) + strlen(sks) + 64); - sprintf(buf, "{\"public_key\":\"%s\",\"secret_key\":\"%s\"}", pks, sks); - OQS_SIG_free(sig); - return el_wrap_str(buf); -} - -el_val_t pq_sign(el_val_t secret_key_hex, el_val_t message) { - size_t sk_len = 0; - unsigned char* sk = el_hex_decode(EL_CSTR(secret_key_hex), &sk_len); - if (!sk) return pq_error("invalid hex in secret_key"); - - OQS_SIG* sig = OQS_SIG_new(EL_DILITHIUM_ALG); - if (!sig) return pq_error("OQS_SIG_new(dilithium-3) failed"); - if (sk_len != sig->length_secret_key) { - OQS_SIG_free(sig); - return pq_error("secret_key length mismatch for dilithium-3"); - } - - const char* msg = EL_CSTR(message); - size_t msg_len = el_input_len(msg); - unsigned char* signature = (unsigned char*)malloc(sig->length_signature); - size_t signature_len = sig->length_signature; - if (!signature) { OQS_SIG_free(sig); return pq_error("oom"); } - - if (OQS_SIG_sign(sig, signature, &signature_len, - (const unsigned char*)(msg ? msg : ""), msg_len, sk) != OQS_SUCCESS) { - free(signature); OQS_SIG_free(sig); - return pq_error("dilithium-3 sign failed"); - } - el_val_t sig_hex = el_hex_encode(signature, signature_len); - free(signature); OQS_SIG_free(sig); - return sig_hex; -} - -el_val_t pq_verify(el_val_t public_key_hex, el_val_t message, el_val_t signature_hex) { - size_t pk_len = 0, sig_len = 0; - unsigned char* pk = el_hex_decode(EL_CSTR(public_key_hex), &pk_len); - unsigned char* signature = el_hex_decode(EL_CSTR(signature_hex), &sig_len); - if (!pk || !signature) return EL_INT(0); - - OQS_SIG* sig = OQS_SIG_new(EL_DILITHIUM_ALG); - if (!sig) return EL_INT(0); - if (pk_len != sig->length_public_key) { OQS_SIG_free(sig); return EL_INT(0); } - - const char* msg = EL_CSTR(message); - size_t msg_len = el_input_len(msg); - OQS_STATUS rc = OQS_SIG_verify(sig, - (const unsigned char*)(msg ? msg : ""), msg_len, - signature, sig_len, pk); - OQS_SIG_free(sig); - return (rc == OQS_SUCCESS) ? EL_INT(1) : EL_INT(0); -} - -/* ─── Kyber-768 / ML-KEM-768 KEM ──────────────────────────────────────── - * - * NIST FIPS 203 standardized CRYSTALS-Kyber as ML-KEM. ML-KEM-768 is the - * FIPS form of what we historically called Kyber-768. Same situation as - * Dilithium → ML-DSA: prefer the standardized constant, fall back to the - * legacy name. liboqs 0.15.0 still exposes OQS_KEM_alg_kyber_768; the - * algorithm is identical at the wire level to ML-KEM-768 except for FIPS - * domain-separation tweaks, so the two ciphertexts/keys are NOT - * cross-compatible. Pin the constant for archival material. */ - -#if defined(OQS_KEM_alg_ml_kem_768) -# define EL_KYBER_ALG OQS_KEM_alg_ml_kem_768 -#elif defined(OQS_KEM_alg_kyber_768) -# define EL_KYBER_ALG OQS_KEM_alg_kyber_768 -#else -# define EL_KYBER_ALG "ML-KEM-768" -#endif - -el_val_t pq_kem_keygen(void) { - OQS_KEM* kem = OQS_KEM_new(EL_KYBER_ALG); - if (!kem) return pq_error("OQS_KEM_new(kyber-768) failed"); - unsigned char* pk = (unsigned char*)malloc(kem->length_public_key); - unsigned char* sk = (unsigned char*)malloc(kem->length_secret_key); - if (!pk || !sk) { free(pk); free(sk); OQS_KEM_free(kem); return pq_error("oom"); } - if (OQS_KEM_keypair(kem, pk, sk) != OQS_SUCCESS) { - free(pk); free(sk); OQS_KEM_free(kem); - return pq_error("kyber-768 keypair generation failed"); - } - el_val_t pk_hex = el_hex_encode(pk, kem->length_public_key); - el_val_t sk_hex = el_hex_encode(sk, kem->length_secret_key); - OQS_MEM_secure_free(sk, kem->length_secret_key); - free(pk); - - const char* pks = EL_CSTR(pk_hex); - const char* sks = EL_CSTR(sk_hex); - char* buf = el_strbuf(strlen(pks) + strlen(sks) + 64); - sprintf(buf, "{\"public_key\":\"%s\",\"secret_key\":\"%s\"}", pks, sks); - OQS_KEM_free(kem); - return el_wrap_str(buf); -} - -el_val_t pq_kem_encaps(el_val_t public_key_hex) { - size_t pk_len = 0; - unsigned char* pk = el_hex_decode(EL_CSTR(public_key_hex), &pk_len); - if (!pk) return pq_error("invalid hex in public_key"); - - OQS_KEM* kem = OQS_KEM_new(EL_KYBER_ALG); - if (!kem) return pq_error("OQS_KEM_new(kyber-768) failed"); - if (pk_len != kem->length_public_key) { - OQS_KEM_free(kem); - return pq_error("public_key length mismatch for kyber-768"); - } - unsigned char* ct = (unsigned char*)malloc(kem->length_ciphertext); - unsigned char* ss = (unsigned char*)malloc(kem->length_shared_secret); - if (!ct || !ss) { free(ct); free(ss); OQS_KEM_free(kem); return pq_error("oom"); } - if (OQS_KEM_encaps(kem, ct, ss, pk) != OQS_SUCCESS) { - free(ct); free(ss); OQS_KEM_free(kem); - return pq_error("kyber-768 encapsulation failed"); - } - el_val_t ct_hex = el_hex_encode(ct, kem->length_ciphertext); - el_val_t ss_hex = el_hex_encode(ss, kem->length_shared_secret); - free(ct); - OQS_MEM_secure_free(ss, kem->length_shared_secret); - - const char* cts = EL_CSTR(ct_hex); - const char* sss = EL_CSTR(ss_hex); - char* buf = el_strbuf(strlen(cts) + strlen(sss) + 64); - sprintf(buf, "{\"ciphertext\":\"%s\",\"shared_secret\":\"%s\"}", cts, sss); - OQS_KEM_free(kem); - return el_wrap_str(buf); -} - -el_val_t pq_kem_decaps(el_val_t secret_key_hex, el_val_t ciphertext_hex) { - size_t sk_len = 0, ct_len = 0; - unsigned char* sk = el_hex_decode(EL_CSTR(secret_key_hex), &sk_len); - unsigned char* ct = el_hex_decode(EL_CSTR(ciphertext_hex), &ct_len); - if (!sk || !ct) return pq_error("invalid hex in inputs"); - - OQS_KEM* kem = OQS_KEM_new(EL_KYBER_ALG); - if (!kem) return pq_error("OQS_KEM_new(kyber-768) failed"); - if (sk_len != kem->length_secret_key || ct_len != kem->length_ciphertext) { - OQS_KEM_free(kem); - return pq_error("input length mismatch for kyber-768"); - } - unsigned char* ss = (unsigned char*)malloc(kem->length_shared_secret); - if (!ss) { OQS_KEM_free(kem); return pq_error("oom"); } - /* Kyber is IND-CCA via Fujisaki-Okamoto: decaps always returns *some* - * shared_secret even on tampered ciphertext (an implicit-rejection value - * derived from sk). Protocols MUST confirm the shared_secret matches via - * a subsequent step (e.g. AEAD tag, key-confirmation MAC) — do not - * assume decaps success implies authenticity. */ - if (OQS_KEM_decaps(kem, ss, ct, sk) != OQS_SUCCESS) { - free(ss); OQS_KEM_free(kem); - return pq_error("kyber-768 decapsulation failed"); - } - el_val_t ss_hex = el_hex_encode(ss, kem->length_shared_secret); - OQS_MEM_secure_free(ss, kem->length_shared_secret); - OQS_KEM_free(kem); - return ss_hex; -} - -/* ─── Hybrid handshake (X25519 + Kyber-768, HKDF-SHA256 combined) ─────── */ - -#if !EL_HAVE_OPENSSL - -el_val_t pq_hybrid_keygen(void) { - return pq_error("hybrid handshake requires OpenSSL (X25519); rebuild with -lcrypto"); -} -el_val_t pq_hybrid_handshake(el_val_t pub) { - (void)pub; - return pq_error("hybrid handshake requires OpenSSL (X25519); rebuild with -lcrypto"); -} - -#else /* EL_HAVE_OPENSSL */ - -/* HKDF-SHA256 (RFC 5869) — Extract+Expand. Reuses the inline HMAC-SHA256 - * already in this file. Empty salt → 32 zero bytes per the RFC. */ -static void el_hkdf_sha256(const unsigned char* salt, size_t salt_len, - const unsigned char* ikm, size_t ikm_len, - const unsigned char* info, size_t info_len, - unsigned char* out, size_t out_len) { - unsigned char zero_salt[32] = {0}; - if (salt_len == 0) { salt = zero_salt; salt_len = 32; } - unsigned char prk[32]; - el_hmac_sha256(salt, salt_len, ikm, ikm_len, prk); - - unsigned char t[32]; - size_t produced = 0; - unsigned char counter = 1; - unsigned char* buf = (unsigned char*)malloc(32 + info_len + 1); - if (!buf) { fputs("el_runtime: hkdf oom\n", stderr); return; } - while (produced < out_len) { - size_t off = 0; - if (counter > 1) { memcpy(buf, t, 32); off = 32; } - if (info && info_len) { memcpy(buf + off, info, info_len); off += info_len; } - buf[off++] = counter; - el_hmac_sha256(prk, 32, buf, off, t); - size_t chunk = (out_len - produced > 32) ? 32 : (out_len - produced); - memcpy(out + produced, t, chunk); - produced += chunk; - counter++; - } - free(buf); -} - -/* X25519 keygen via OpenSSL EVP. Returns 1 on success. - * Fills pk[32] and sk[32] (raw X25519 byte strings, no DER wrapper). */ -static int el_x25519_keygen(unsigned char pk[32], unsigned char sk[32]) { - EVP_PKEY_CTX* pctx = EVP_PKEY_CTX_new_id(EVP_PKEY_X25519, NULL); - if (!pctx) return 0; - if (EVP_PKEY_keygen_init(pctx) != 1) { EVP_PKEY_CTX_free(pctx); return 0; } - EVP_PKEY* key = NULL; - if (EVP_PKEY_keygen(pctx, &key) != 1) { EVP_PKEY_CTX_free(pctx); return 0; } - EVP_PKEY_CTX_free(pctx); - - size_t plen = 32, slen = 32; - if (EVP_PKEY_get_raw_public_key (key, pk, &plen) != 1 || plen != 32) { - EVP_PKEY_free(key); return 0; - } - if (EVP_PKEY_get_raw_private_key(key, sk, &slen) != 1 || slen != 32) { - EVP_PKEY_free(key); return 0; - } - EVP_PKEY_free(key); - return 1; -} - -/* X25519 ECDH: derive 32-byte shared secret from local sk and remote pk. */ -static int el_x25519_derive(const unsigned char sk[32], const unsigned char rpk[32], - unsigned char ss[32]) { - EVP_PKEY* my = EVP_PKEY_new_raw_private_key(EVP_PKEY_X25519, NULL, sk, 32); - EVP_PKEY* rem = EVP_PKEY_new_raw_public_key (EVP_PKEY_X25519, NULL, rpk, 32); - if (!my || !rem) { EVP_PKEY_free(my); EVP_PKEY_free(rem); return 0; } - EVP_PKEY_CTX* dctx = EVP_PKEY_CTX_new(my, NULL); - if (!dctx) { EVP_PKEY_free(my); EVP_PKEY_free(rem); return 0; } - int ok = 0; - size_t out_len = 32; - if (EVP_PKEY_derive_init(dctx) == 1 && - EVP_PKEY_derive_set_peer(dctx, rem) == 1 && - EVP_PKEY_derive(dctx, ss, &out_len) == 1 && - out_len == 32) ok = 1; - EVP_PKEY_CTX_free(dctx); - EVP_PKEY_free(my); - EVP_PKEY_free(rem); - return ok; -} - -/* Hybrid wire layout (binary form, before hex encode): - * public_key = x25519_pub (32) || kyber_pub (1184) → 1216 bytes - * secret_key = x25519_sec (32) || kyber_sec (2400) → 2432 bytes - * ciphertext = ephem_x25519_pub (32) || kyber_ct (1088) → 1120 bytes - * shared_secret = HKDF-SHA256(x25519_ss || kyber_ss, info="el-pq-hybrid-v1", 32 bytes) - * The keygen result also exposes the four component hex fields for callers - * that prefer to handle the legs independently. */ - -el_val_t pq_hybrid_keygen(void) { - OQS_KEM* kem = OQS_KEM_new(EL_KYBER_ALG); - if (!kem) return pq_error("OQS_KEM_new(kyber-768) failed"); - - unsigned char xpk[32], xsk[32]; - if (!el_x25519_keygen(xpk, xsk)) { - OQS_KEM_free(kem); - return pq_error("X25519 keygen failed"); - } - - unsigned char* kpk = (unsigned char*)malloc(kem->length_public_key); - unsigned char* ksk = (unsigned char*)malloc(kem->length_secret_key); - if (!kpk || !ksk) { free(kpk); free(ksk); OQS_KEM_free(kem); return pq_error("oom"); } - if (OQS_KEM_keypair(kem, kpk, ksk) != OQS_SUCCESS) { - free(kpk); free(ksk); OQS_KEM_free(kem); - return pq_error("kyber-768 keypair generation failed"); - } - - size_t pub_len = 32 + kem->length_public_key; - size_t sec_len = 32 + kem->length_secret_key; - unsigned char* pub_buf = (unsigned char*)malloc(pub_len); - unsigned char* sec_buf = (unsigned char*)malloc(sec_len); - if (!pub_buf || !sec_buf) { - free(pub_buf); free(sec_buf); free(kpk); - OQS_MEM_secure_free(ksk, kem->length_secret_key); - OQS_KEM_free(kem); return pq_error("oom"); - } - memcpy(pub_buf, xpk, 32); memcpy(pub_buf + 32, kpk, kem->length_public_key); - memcpy(sec_buf, xsk, 32); memcpy(sec_buf + 32, ksk, kem->length_secret_key); - - el_val_t x_pub_hex = el_hex_encode(xpk, 32); - el_val_t x_sec_hex = el_hex_encode(xsk, 32); - el_val_t k_pub_hex = el_hex_encode(kpk, kem->length_public_key); - el_val_t k_sec_hex = el_hex_encode(ksk, kem->length_secret_key); - el_val_t pub_hex = el_hex_encode(pub_buf, pub_len); - el_val_t sec_hex = el_hex_encode(sec_buf, sec_len); - - OQS_MEM_secure_free(ksk, kem->length_secret_key); - free(kpk); free(pub_buf); free(sec_buf); - OQS_KEM_free(kem); - memset(xsk, 0, 32); /* best-effort wipe of stack copy */ - - const char* xph = EL_CSTR(x_pub_hex); - const char* xsh = EL_CSTR(x_sec_hex); - const char* kph = EL_CSTR(k_pub_hex); - const char* ksh = EL_CSTR(k_sec_hex); - const char* pubh = EL_CSTR(pub_hex); - const char* sech = EL_CSTR(sec_hex); - - char* buf = el_strbuf(strlen(xph) + strlen(xsh) + strlen(kph) + strlen(ksh) - + strlen(pubh) + strlen(sech) + 256); - sprintf(buf, - "{\"x25519_pub\":\"%s\",\"x25519_sec\":\"%s\"," - "\"kyber_pub\":\"%s\",\"kyber_sec\":\"%s\"," - "\"public_key\":\"%s\",\"secret_key\":\"%s\"}", - xph, xsh, kph, ksh, pubh, sech); - return el_wrap_str(buf); -} - -/* Initiator-side handshake. Caller supplies the responder's combined public - * key (x25519_pub || kyber_pub, hex-encoded). The runtime: - * 1. Generates an ephemeral X25519 keypair, runs ECDH against the - * responder's static x25519_pub. - * 2. Runs Kyber-768 encaps against the responder's kyber_pub → kyber_ct, - * kyber_ss. - * 3. Combined shared = HKDF-SHA256(salt="", ikm = x25519_ss || kyber_ss, - * info = "el-pq-hybrid-v1", L = 32). - * 4. Returns combined ciphertext (= ephemeral_x25519_pub || kyber_ct) and - * the derived shared_secret. - * - * Responder side composition (intentionally not a separate runtime fn — - * trivial to express in El given pq_kem_decaps + a future x25519_derive - * primitive): split the ciphertext into ephem_xpk (32) and kyber_ct, run - * X25519(static_xsk, ephem_xpk) and pq_kem_decaps(static_kyber_sk, kyber_ct), - * then HKDF-SHA256 with the same salt/info to recover the same shared_secret. - * If a separate x25519 entry point becomes valuable, add `pq_hybrid_open` - * here taking (secret_key_combined, ciphertext_combined). */ -el_val_t pq_hybrid_handshake(el_val_t remote_pub_combined) { - size_t pub_len = 0; - unsigned char* rpub = el_hex_decode(EL_CSTR(remote_pub_combined), &pub_len); - if (!rpub) return pq_error("invalid hex in remote_pub_combined"); - - OQS_KEM* kem = OQS_KEM_new(EL_KYBER_ALG); - if (!kem) return pq_error("OQS_KEM_new(kyber-768) failed"); - if (pub_len != 32 + kem->length_public_key) { - OQS_KEM_free(kem); - return pq_error("remote_pub_combined length mismatch (expected x25519_pub || kyber_pub)"); - } - - unsigned char e_xpk[32], e_xsk[32], x_ss[32]; - if (!el_x25519_keygen(e_xpk, e_xsk)) { - OQS_KEM_free(kem); - return pq_error("X25519 ephemeral keygen failed"); - } - if (!el_x25519_derive(e_xsk, rpub, x_ss)) { - memset(e_xsk, 0, 32); - OQS_KEM_free(kem); - return pq_error("X25519 derive failed"); - } - memset(e_xsk, 0, 32); /* ephemeral; not needed after derive */ - - unsigned char* k_ct = (unsigned char*)malloc(kem->length_ciphertext); - unsigned char* k_ss = (unsigned char*)malloc(kem->length_shared_secret); - if (!k_ct || !k_ss) { - free(k_ct); free(k_ss); OQS_KEM_free(kem); - return pq_error("oom"); - } - if (OQS_KEM_encaps(kem, k_ct, k_ss, rpub + 32) != OQS_SUCCESS) { - free(k_ct); free(k_ss); OQS_KEM_free(kem); - return pq_error("kyber-768 encapsulation failed"); - } - - /* HKDF combine: ikm = x_ss || k_ss. */ - size_t ikm_len = 32 + kem->length_shared_secret; - unsigned char* ikm = (unsigned char*)malloc(ikm_len); - if (!ikm) { - free(k_ct); OQS_MEM_secure_free(k_ss, kem->length_shared_secret); - OQS_KEM_free(kem); - return pq_error("oom"); - } - memcpy(ikm, x_ss, 32); - memcpy(ikm + 32, k_ss, kem->length_shared_secret); - unsigned char combined[32]; - static const char info_str[] = "el-pq-hybrid-v1"; - el_hkdf_sha256(NULL, 0, ikm, ikm_len, - (const unsigned char*)info_str, sizeof(info_str) - 1, - combined, 32); - - memset(x_ss, 0, 32); - OQS_MEM_secure_free(k_ss, kem->length_shared_secret); - OQS_MEM_secure_free(ikm, ikm_len); - - /* Combined ciphertext = ephemeral_x25519_pub || kyber_ct. */ - size_t ct_len = 32 + kem->length_ciphertext; - unsigned char* combined_ct = (unsigned char*)malloc(ct_len); - if (!combined_ct) { free(k_ct); OQS_KEM_free(kem); return pq_error("oom"); } - memcpy(combined_ct, e_xpk, 32); - memcpy(combined_ct + 32, k_ct, kem->length_ciphertext); - free(k_ct); - OQS_KEM_free(kem); - - el_val_t ct_hex = el_hex_encode(combined_ct, ct_len); - el_val_t ss_hex = el_hex_encode(combined, 32); - free(combined_ct); - memset(combined, 0, 32); - - const char* cts = EL_CSTR(ct_hex); - const char* sss = EL_CSTR(ss_hex); - char* buf = el_strbuf(strlen(cts) + strlen(sss) + 64); - sprintf(buf, "{\"ciphertext\":\"%s\",\"shared_secret\":\"%s\"}", cts, sss); - return el_wrap_str(buf); -} - -#endif /* EL_HAVE_OPENSSL */ -#endif /* EL_HAVE_LIBOQS */ - -/* ─── AEAD: AES-256-GCM ──────────────────────────────────────────────────── - * - * Symmetric authenticated encryption used to wrap envelopes once a shared - * secret has been derived from the KEM (Kyber-768 / hybrid). The El surface - * is intentionally narrow: - * - * aead_encrypt(key_hex, plaintext) - * → {"nonce":"<24 hex>","ciphertext":"<...hex including 16-byte tag>"} - * - * aead_decrypt(key_hex, nonce_hex, ciphertext_hex) - * → plaintext String, or "" on auth failure / malformed input - * - * Conventions: - * - key_hex must decode to exactly 32 bytes (AES-256). Callers that hold - * a longer KEM shared_secret should normalize via SHA3-256(ss) → 32 bytes - * before passing it in. (Kyber-768's shared_secret is already 32 bytes, - * but keeping this contract explicit lets the El side be agnostic.) - * - nonce is a fresh 12-byte random value drawn from the OS CSPRNG. Caller - * never picks the nonce — eliminates the GCM nonce-reuse footgun entirely. - * - tag is the standard 16 bytes, appended to ciphertext per RFC 5116. - * `ciphertext` field is therefore (plaintext_len + 16) bytes, hex-encoded. - * - No associated data (AAD). If we later need bound metadata, add a - * length-prefixed AAD argument and bump the envelope version tag. - * - * Failure mode: - * aead_encrypt returns http_error_json(...) on input/system failure. - * aead_decrypt returns the empty string on ANY failure (including auth-tag - * mismatch). Callers MUST check for "" before using the result. */ - -#if !__has_include() - -el_val_t aead_encrypt(el_val_t key_hex, el_val_t plaintext) { - (void)key_hex; (void)plaintext; - return http_error_json("aead_encrypt requires OpenSSL (libcrypto); rebuild with -lcrypto"); -} -el_val_t aead_decrypt(el_val_t key_hex, el_val_t nonce_hex, el_val_t ciphertext_hex) { - (void)key_hex; (void)nonce_hex; (void)ciphertext_hex; - return el_wrap_str(el_strdup("")); -} - -#else /* OpenSSL available */ - -#include -#include - -el_val_t aead_encrypt(el_val_t key_hex, el_val_t plaintext) { - size_t key_len = 0; - unsigned char* key = el_hex_decode(EL_CSTR(key_hex), &key_len); - if (!key) return http_error_json("invalid hex in key"); - if (key_len != 32) return http_error_json("aead key must be 32 bytes (64 hex chars) for AES-256-GCM"); - - const char* pt = EL_CSTR(plaintext); - size_t pt_len = el_input_len(pt); - if (!pt) pt = ""; - - unsigned char nonce[12]; - if (RAND_bytes(nonce, 12) != 1) return http_error_json("OS CSPRNG failed (RAND_bytes)"); - - EVP_CIPHER_CTX* ctx = EVP_CIPHER_CTX_new(); - if (!ctx) return http_error_json("EVP_CIPHER_CTX_new failed"); - - if (EVP_EncryptInit_ex(ctx, EVP_aes_256_gcm(), NULL, NULL, NULL) != 1) { - EVP_CIPHER_CTX_free(ctx); return http_error_json("aes-256-gcm init failed"); - } - if (EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_GCM_SET_IVLEN, 12, NULL) != 1) { - EVP_CIPHER_CTX_free(ctx); return http_error_json("set ivlen failed"); - } - if (EVP_EncryptInit_ex(ctx, NULL, NULL, key, nonce) != 1) { - EVP_CIPHER_CTX_free(ctx); return http_error_json("aes-256-gcm key/iv init failed"); - } - - /* GCM ciphertext is the same length as plaintext; we append a 16-byte - * authentication tag for AEAD semantics. Allocate plaintext_len + 16. */ - unsigned char* ct = (unsigned char*)malloc(pt_len + 16); - if (!ct) { EVP_CIPHER_CTX_free(ctx); return http_error_json("oom"); } - int outlen = 0, total = 0; - if (EVP_EncryptUpdate(ctx, ct, &outlen, (const unsigned char*)pt, (int)pt_len) != 1) { - free(ct); EVP_CIPHER_CTX_free(ctx); return http_error_json("aes-256-gcm update failed"); - } - total += outlen; - if (EVP_EncryptFinal_ex(ctx, ct + total, &outlen) != 1) { - free(ct); EVP_CIPHER_CTX_free(ctx); return http_error_json("aes-256-gcm final failed"); - } - total += outlen; - if (EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_GCM_GET_TAG, 16, ct + total) != 1) { - free(ct); EVP_CIPHER_CTX_free(ctx); return http_error_json("aes-256-gcm get tag failed"); - } - EVP_CIPHER_CTX_free(ctx); - - el_val_t nonce_hex_v = el_hex_encode(nonce, 12); - el_val_t ct_hex_v = el_hex_encode(ct, (size_t)total + 16); - free(ct); - - const char* nh = EL_CSTR(nonce_hex_v); - const char* ch = EL_CSTR(ct_hex_v); - char* buf = el_strbuf(strlen(nh) + strlen(ch) + 48); - sprintf(buf, "{\"nonce\":\"%s\",\"ciphertext\":\"%s\"}", nh, ch); - return el_wrap_str(buf); -} - -el_val_t aead_decrypt(el_val_t key_hex, el_val_t nonce_hex, el_val_t ciphertext_hex) { - size_t key_len = 0, nonce_len = 0, ct_len = 0; - unsigned char* key = el_hex_decode(EL_CSTR(key_hex), &key_len); - unsigned char* nonce = el_hex_decode(EL_CSTR(nonce_hex), &nonce_len); - unsigned char* ct = el_hex_decode(EL_CSTR(ciphertext_hex), &ct_len); - if (!key || !nonce || !ct) return el_wrap_str(el_strdup("")); - if (key_len != 32 || nonce_len != 12) return el_wrap_str(el_strdup("")); - if (ct_len < 16) return el_wrap_str(el_strdup("")); - - size_t body_len = ct_len - 16; - const unsigned char* tag = ct + body_len; - - EVP_CIPHER_CTX* ctx = EVP_CIPHER_CTX_new(); - if (!ctx) return el_wrap_str(el_strdup("")); - - if (EVP_DecryptInit_ex(ctx, EVP_aes_256_gcm(), NULL, NULL, NULL) != 1 || - EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_GCM_SET_IVLEN, 12, NULL) != 1 || - EVP_DecryptInit_ex(ctx, NULL, NULL, key, nonce) != 1) { - EVP_CIPHER_CTX_free(ctx); return el_wrap_str(el_strdup("")); - } - - unsigned char* pt = (unsigned char*)malloc(body_len + 1); - if (!pt) { EVP_CIPHER_CTX_free(ctx); return el_wrap_str(el_strdup("")); } - int outlen = 0, total = 0; - if (EVP_DecryptUpdate(ctx, pt, &outlen, ct, (int)body_len) != 1) { - free(pt); EVP_CIPHER_CTX_free(ctx); return el_wrap_str(el_strdup("")); - } - total += outlen; - /* Set expected tag before final — GCM's final step is where auth happens. */ - if (EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_GCM_SET_TAG, 16, (void*)tag) != 1) { - free(pt); EVP_CIPHER_CTX_free(ctx); return el_wrap_str(el_strdup("")); - } - int rc = EVP_DecryptFinal_ex(ctx, pt + total, &outlen); - EVP_CIPHER_CTX_free(ctx); - if (rc != 1) { - /* Auth failure or padding/length mismatch. Return empty so callers - * cannot accidentally treat tampered ciphertext as a valid message. */ - free(pt); - return el_wrap_str(el_strdup("")); - } - total += outlen; - pt[total] = '\0'; - - /* Copy into the el arena so the caller-visible string outlives this fn. */ - char* out = el_strbuf((size_t)total); - memcpy(out, pt, (size_t)total); - out[total] = '\0'; - free(pt); - return el_wrap_str(out); -} - -#endif /* __has_include() */ - -/* ──────────────────────────────────────────────────────────────────────────── - * OTLP/HTTP observability — logs, traces, metrics - * - * Design goals: - * - Zero blocking on the request path. Producers append to in-memory - * ring buffers; a single worker thread flushes to the OTLP endpoint. - * - Drop-on-failure semantics. If the endpoint is unreachable or slow, - * we drop telemetry rather than back-pressure into the request handler. - * - Best-effort serialization. Each record is pre-serialized as JSON when - * the El program calls the primitive; the worker just batches. - * - Configuration via env vars: - * OTLP_ENDPOINT e.g. https://alloy.neuralplatform.ai:4318 - * OTEL_SERVICE_NAME e.g. neuron-web (default: argv[0] basename) - * OTEL_SERVICE_VERSION (default: "0.0.0") - * OTEL_RESOURCE_ATTRS comma-sep k=v pairs (optional) - * - * Wire format: OTLP/HTTP JSON. Three endpoints: - * POST {endpoint}/v1/logs — log records - * POST {endpoint}/v1/traces — spans - * POST {endpoint}/v1/metrics — counter/gauge points - * - * El programs see four primitives: - * trace_span_start(name) -> SpanHandle (just a string id) - * trace_span_end(handle) (computes duration, queues) - * emit_log(level, msg, fields_json) (queues a log record) - * emit_metric(name, value, tags_json) (queues a counter increment) - * ──────────────────────────────────────────────────────────────────────────── - */ - -#define OTLP_BUF_CAP 4096 /* per-buffer ring size */ -#define OTLP_FLUSH_MS 2000 /* flush every 2s */ -#define OTLP_BATCH_MAX 200 /* up to 200 records per POST */ - -typedef struct { - char* data; /* malloc'd JSON fragment for this record */ -} OtlpRec; - -typedef struct { - OtlpRec ring[OTLP_BUF_CAP]; - size_t head; /* next write slot */ - size_t tail; /* next read slot */ - pthread_mutex_t mu; -} OtlpQueue; - -static OtlpQueue _otlp_logs = { .mu = PTHREAD_MUTEX_INITIALIZER }; -static OtlpQueue _otlp_traces = { .mu = PTHREAD_MUTEX_INITIALIZER }; -static OtlpQueue _otlp_metrics = { .mu = PTHREAD_MUTEX_INITIALIZER }; - -static char* _otlp_endpoint = NULL; /* e.g. https://alloy.neuralplatform.ai:4318 */ -static char* _otlp_service_name = NULL; -static char* _otlp_service_version = NULL; -static int _otlp_initialized = 0; -static pthread_t _otlp_worker_thread; - -/* enqueue — returns 1 if accepted, 0 if dropped (full buffer or no endpoint) */ -static int otlp_enqueue(OtlpQueue* q, const char* json) { - if (!_otlp_endpoint || !json) return 0; - pthread_mutex_lock(&q->mu); - size_t next_head = (q->head + 1) % OTLP_BUF_CAP; - if (next_head == q->tail) { - /* buffer full — drop oldest */ - free(q->ring[q->tail].data); - q->ring[q->tail].data = NULL; - q->tail = (q->tail + 1) % OTLP_BUF_CAP; - } - q->ring[q->head].data = strdup(json); - q->head = next_head; - pthread_mutex_unlock(&q->mu); - return 1; -} - -/* drain — copies up to OTLP_BATCH_MAX items into a comma-joined string, - * caller must free the result. Returns NULL if queue is empty. */ -static char* otlp_drain(OtlpQueue* q) { - pthread_mutex_lock(&q->mu); - if (q->head == q->tail) { pthread_mutex_unlock(&q->mu); return NULL; } - /* compute total length */ - size_t total = 0, count = 0; - size_t i = q->tail; - while (i != q->head && count < OTLP_BATCH_MAX) { - if (q->ring[i].data) total += strlen(q->ring[i].data) + 1; /* +1 for comma */ - i = (i + 1) % OTLP_BUF_CAP; - count++; - } - char* out = malloc(total + 4); - if (!out) { pthread_mutex_unlock(&q->mu); return NULL; } - out[0] = '\0'; - size_t off = 0; - i = q->tail; - count = 0; - while (i != q->head && count < OTLP_BATCH_MAX) { - if (q->ring[i].data) { - size_t l = strlen(q->ring[i].data); - if (off > 0) { out[off++] = ','; } - memcpy(out + off, q->ring[i].data, l); - off += l; - free(q->ring[i].data); - q->ring[i].data = NULL; - } - i = (i + 1) % OTLP_BUF_CAP; - count++; - } - out[off] = '\0'; - q->tail = i; - pthread_mutex_unlock(&q->mu); - return out; -} - -/* Build resource block once (service.name, service.version, host.name) */ -static char* otlp_resource_block(void) { - static char cached[1024]; - static int built = 0; - if (built) return cached; - char host[256] = "unknown"; - gethostname(host, sizeof(host) - 1); - snprintf(cached, sizeof(cached), - "{\"attributes\":[" - "{\"key\":\"service.name\",\"value\":{\"stringValue\":\"%s\"}}," - "{\"key\":\"service.version\",\"value\":{\"stringValue\":\"%s\"}}," - "{\"key\":\"host.name\",\"value\":{\"stringValue\":\"%s\"}}" - "]}", - _otlp_service_name ? _otlp_service_name : "el-app", - _otlp_service_version ? _otlp_service_version : "0.0.0", - host); - built = 1; - return cached; -} - -/* Best-effort POST. Drops on any error. */ -static void otlp_post(const char* path, const char* body) { - if (!_otlp_endpoint || !body || !*body) return; - char url[1024]; - snprintf(url, sizeof(url), "%s%s", _otlp_endpoint, path); - CURL* c = curl_easy_init(); - if (!c) return; - struct curl_slist* h = NULL; - h = curl_slist_append(h, "Content-Type: application/json"); - curl_easy_setopt(c, CURLOPT_URL, url); - curl_easy_setopt(c, CURLOPT_POST, 1L); - curl_easy_setopt(c, CURLOPT_POSTFIELDS, body); - curl_easy_setopt(c, CURLOPT_POSTFIELDSIZE, (long)strlen(body)); - curl_easy_setopt(c, CURLOPT_HTTPHEADER, h); - curl_easy_setopt(c, CURLOPT_TIMEOUT_MS, 3000L); - curl_easy_setopt(c, CURLOPT_NOSIGNAL, 1L); - curl_easy_setopt(c, CURLOPT_WRITEFUNCTION, NULL); /* discard response */ - curl_easy_perform(c); - curl_slist_free_all(h); - curl_easy_cleanup(c); -} - -/* Flush worker — runs forever until process exits */ -static void* otlp_worker(void* arg) { - (void)arg; - while (1) { - struct timespec ts = { OTLP_FLUSH_MS / 1000, (OTLP_FLUSH_MS % 1000) * 1000000L }; - nanosleep(&ts, NULL); - - char* logs = otlp_drain(&_otlp_logs); - if (logs && *logs) { - char body[OTLP_BUF_CAP * 8]; - int n = snprintf(body, sizeof(body), - "{\"resourceLogs\":[{\"resource\":%s," - "\"scopeLogs\":[{\"scope\":{\"name\":\"el-runtime\"}," - "\"logRecords\":[%s]}]}]}", - otlp_resource_block(), logs); - if (n > 0 && n < (int)sizeof(body)) otlp_post("/v1/logs", body); - } - free(logs); - - char* traces = otlp_drain(&_otlp_traces); - if (traces && *traces) { - char body[OTLP_BUF_CAP * 8]; - int n = snprintf(body, sizeof(body), - "{\"resourceSpans\":[{\"resource\":%s," - "\"scopeSpans\":[{\"scope\":{\"name\":\"el-runtime\"}," - "\"spans\":[%s]}]}]}", - otlp_resource_block(), traces); - if (n > 0 && n < (int)sizeof(body)) otlp_post("/v1/traces", body); - } - free(traces); - - char* metrics = otlp_drain(&_otlp_metrics); - if (metrics && *metrics) { - char body[OTLP_BUF_CAP * 8]; - int n = snprintf(body, sizeof(body), - "{\"resourceMetrics\":[{\"resource\":%s," - "\"scopeMetrics\":[{\"scope\":{\"name\":\"el-runtime\"}," - "\"metrics\":[%s]}]}]}", - otlp_resource_block(), metrics); - if (n > 0 && n < (int)sizeof(body)) otlp_post("/v1/metrics", body); - } - free(metrics); - } - return NULL; -} - -/* Initialize OTLP — called lazily on first emit. Idempotent. */ -static void otlp_lazy_init(void) { - if (_otlp_initialized) return; - static pthread_mutex_t once_mu = PTHREAD_MUTEX_INITIALIZER; - pthread_mutex_lock(&once_mu); - if (_otlp_initialized) { pthread_mutex_unlock(&once_mu); return; } - - const char* ep = getenv("OTLP_ENDPOINT"); - if (!ep || !*ep) { - _otlp_initialized = 1; - pthread_mutex_unlock(&once_mu); - return; - } - _otlp_endpoint = strdup(ep); - /* trim trailing slash */ - size_t l = strlen(_otlp_endpoint); - if (l > 0 && _otlp_endpoint[l - 1] == '/') _otlp_endpoint[l - 1] = '\0'; - - const char* svc = getenv("OTEL_SERVICE_NAME"); - _otlp_service_name = strdup(svc && *svc ? svc : "el-app"); - const char* ver = getenv("OTEL_SERVICE_VERSION"); - _otlp_service_version = strdup(ver && *ver ? ver : "0.0.0"); - - pthread_create(&_otlp_worker_thread, NULL, otlp_worker, NULL); - pthread_detach(_otlp_worker_thread); - _otlp_initialized = 1; - pthread_mutex_unlock(&once_mu); -} - -/* JSON-escape a string into out_buf. Returns chars written (excluding null). */ -static size_t otlp_json_escape(const char* in, char* out, size_t out_cap) { - size_t o = 0; - for (size_t i = 0; in[i] && o + 8 < out_cap; i++) { - unsigned char c = (unsigned char)in[i]; - if (c == '"') { out[o++] = '\\'; out[o++] = '"'; } - else if (c == '\\'){ out[o++] = '\\'; out[o++] = '\\'; } - else if (c == '\n'){ out[o++] = '\\'; out[o++] = 'n'; } - else if (c == '\r'){ out[o++] = '\\'; out[o++] = 'r'; } - else if (c == '\t'){ out[o++] = '\\'; out[o++] = 't'; } - else if (c < 0x20) { o += snprintf(out + o, out_cap - o, "\\u%04x", c); } - else { out[o++] = (char)c; } - } - out[o] = '\0'; - return o; -} - -/* ── Public El primitives ─────────────────────────────────────────────────── */ - -/* emit_log(level, msg, fields_json) — fields_json is a JSON object string or "" */ -el_val_t emit_log(el_val_t level_v, el_val_t msg_v, el_val_t fields_v) { - otlp_lazy_init(); - if (!_otlp_endpoint) return EL_INT(0); - const char* level = EL_CSTR(level_v); if (!level) level = "INFO"; - const char* msg = EL_CSTR(msg_v); if (!msg) msg = ""; - const char* fields = EL_CSTR(fields_v); if (!fields) fields = ""; - /* Map El level names to OTLP severity numbers */ - int sev_num = 9; /* INFO */ - if (strcmp(level, "TRACE") == 0) sev_num = 1; - else if (strcmp(level, "DEBUG") == 0) sev_num = 5; - else if (strcmp(level, "INFO") == 0) sev_num = 9; - else if (strcmp(level, "WARN") == 0 || strcmp(level, "WARNING") == 0) sev_num = 13; - else if (strcmp(level, "ERROR") == 0) sev_num = 17; - else if (strcmp(level, "FATAL") == 0) sev_num = 21; - char esc_msg[2048]; otlp_json_escape(msg, esc_msg, sizeof(esc_msg)); - /* unix nanos */ - struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); - long long now_nano = (long long)ts.tv_sec * 1000000000LL + ts.tv_nsec; - char rec[4096]; - int n = snprintf(rec, sizeof(rec), - "{\"timeUnixNano\":\"%lld\",\"severityNumber\":%d," - "\"severityText\":\"%s\"," - "\"body\":{\"stringValue\":\"%s\"}%s%s}", - now_nano, sev_num, level, esc_msg, - (fields && *fields) ? ",\"attributes\":" : "", - (fields && *fields) ? fields : ""); - if (n > 0 && n < (int)sizeof(rec)) otlp_enqueue(&_otlp_logs, rec); - return EL_INT(1); -} - -/* emit_metric(name, value, tags_json) — Sum (counter) data point. tags_json - * is a JSON array of {key, value} pairs or empty string. */ -el_val_t emit_metric(el_val_t name_v, el_val_t value_v, el_val_t tags_v) { - otlp_lazy_init(); - if (!_otlp_endpoint) return EL_INT(0); - const char* name = EL_CSTR(name_v); if (!name) name = "unknown"; - int64_t val = (int64_t)value_v; - const char* tags = EL_CSTR(tags_v); if (!tags) tags = ""; - char esc_name[256]; otlp_json_escape(name, esc_name, sizeof(esc_name)); - struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); - long long now_nano = (long long)ts.tv_sec * 1000000000LL + ts.tv_nsec; - char rec[4096]; - int n = snprintf(rec, sizeof(rec), - "{\"name\":\"%s\",\"sum\":{\"aggregationTemporality\":2,\"isMonotonic\":true," - "\"dataPoints\":[{\"asInt\":\"%lld\"," - "\"timeUnixNano\":\"%lld\"" - "%s%s}]}}", - esc_name, (long long)val, now_nano, - (tags && *tags) ? ",\"attributes\":" : "", - (tags && *tags) ? tags : ""); - if (n > 0 && n < (int)sizeof(rec)) otlp_enqueue(&_otlp_metrics, rec); - return EL_INT(1); -} - -/* trace_span_start(name) — returns a span handle (string of "traceid:spanid:start_nano:name") */ -el_val_t trace_span_start(el_val_t name_v) { - otlp_lazy_init(); - const char* name = EL_CSTR(name_v); if (!name) name = "span"; - /* generate 16-byte trace id and 8-byte span id */ - static _Thread_local int seeded = 0; - if (!seeded) { srand((unsigned int)(uintptr_t)pthread_self() ^ (unsigned int)time(NULL)); seeded = 1; } - char tid[33], sid[17]; - for (int i = 0; i < 32; i++) tid[i] = "0123456789abcdef"[rand() & 0xF]; - tid[32] = '\0'; - for (int i = 0; i < 16; i++) sid[i] = "0123456789abcdef"[rand() & 0xF]; - sid[16] = '\0'; - struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); - long long now_nano = (long long)ts.tv_sec * 1000000000LL + ts.tv_nsec; - char* handle = malloc(strlen(name) + 80); - if (!handle) return EL_STR(""); - sprintf(handle, "%s:%s:%lld:%s", tid, sid, now_nano, name); - el_arena_track(handle); - return EL_STR(handle); -} - -/* trace_span_end(handle) — emits the span with computed duration */ -el_val_t trace_span_end(el_val_t handle_v) { - otlp_lazy_init(); - if (!_otlp_endpoint) return EL_INT(0); - const char* h = EL_CSTR(handle_v); if (!h) return EL_INT(0); - /* parse "tid:sid:start_nano:name" */ - char tid[64], sid[32], rest[1024]; - long long start_nano = 0; - if (sscanf(h, "%63[^:]:%31[^:]:%lld:%1023[^\n]", tid, sid, &start_nano, rest) != 4) return EL_INT(0); - struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); - long long end_nano = (long long)ts.tv_sec * 1000000000LL + ts.tv_nsec; - char esc_name[1024]; otlp_json_escape(rest, esc_name, sizeof(esc_name)); - char rec[4096]; - int n = snprintf(rec, sizeof(rec), - "{\"traceId\":\"%s\",\"spanId\":\"%s\"," - "\"name\":\"%s\"," - "\"kind\":1," - "\"startTimeUnixNano\":\"%lld\"," - "\"endTimeUnixNano\":\"%lld\"," - "\"status\":{\"code\":1}}", - tid, sid, esc_name, start_nano, end_nano); - if (n > 0 && n < (int)sizeof(rec)) otlp_enqueue(&_otlp_traces, rec); - return EL_INT(1); -} - -/* Convenience: emit a one-shot timed event (emit start+end immediately). - * For El programs that want point events with duration baked in. */ -el_val_t emit_event(el_val_t name_v, el_val_t duration_ms_v) { - otlp_lazy_init(); - if (!_otlp_endpoint) return EL_INT(0); - const char* name = EL_CSTR(name_v); if (!name) name = "event"; - int64_t dur_ms = (int64_t)duration_ms_v; - el_val_t h = trace_span_start(EL_STR((char*)name)); - /* fudge start to be (now - duration) */ - (void)dur_ms; - return trace_span_end(h); -} - -/* ── Threading seed primitives ─────────────────────────────────────────────── - * __thread_create(fn_name, arg) -> Int spawn El fn in a pthread, return tid - * __thread_join(tid) -> String join thread, return result string - * __mutex_new() -> Int allocate a mutex, return handle - * __mutex_lock(m) lock mutex m - * __mutex_unlock(m) unlock mutex m - * - * Every El fn compiles to a global C symbol. __thread_create uses dlsym to - * look up the function by name and run it in a pthread. This means any El fn - * with signature (String) -> String is directly threadable. - */ - -typedef el_val_t (*ElFn1)(el_val_t); - -typedef struct { - ElFn1 fn; - el_val_t arg; - el_val_t result; -} ElThreadArg; - -#define EL_THREAD_MAX 256 - -typedef struct { - pthread_t tid; - ElThreadArg* arg; - int alive; -} ElThread; - -static ElThread _threads[EL_THREAD_MAX]; -static int _thread_count = 0; -static pthread_mutex_t _thread_alloc_mu = PTHREAD_MUTEX_INITIALIZER; - -static void* el_thread_runner(void* raw) { - ElThreadArg* a = (ElThreadArg*)raw; - a->result = a->fn(a->arg); - return NULL; -} - -el_val_t __thread_create(el_val_t fn_name_v, el_val_t arg_v) { - const char* sym = EL_CSTR(fn_name_v); - if (!sym || !*sym) return EL_INT(-1); - void* p = dlsym(RTLD_DEFAULT, sym); - if (!p) { - fprintf(stderr, "[__thread_create] symbol not found: %s\n", sym); - return EL_INT(-1); - } - ElThreadArg* a = (ElThreadArg*)malloc(sizeof(ElThreadArg)); - if (!a) return EL_INT(-1); - a->fn = (ElFn1)p; - a->arg = arg_v; - a->result = EL_STR(""); - - pthread_mutex_lock(&_thread_alloc_mu); - if (_thread_count >= EL_THREAD_MAX) { - pthread_mutex_unlock(&_thread_alloc_mu); - free(a); - fprintf(stderr, "[__thread_create] thread table full\n"); - return EL_INT(-1); - } - int slot = _thread_count++; - _threads[slot].arg = a; - _threads[slot].alive = 1; - pthread_mutex_unlock(&_thread_alloc_mu); - - if (pthread_create(&_threads[slot].tid, NULL, el_thread_runner, a) != 0) { - pthread_mutex_lock(&_thread_alloc_mu); - _thread_count--; - pthread_mutex_unlock(&_thread_alloc_mu); - free(a); - return EL_INT(-1); - } - return EL_INT(slot); -} - -el_val_t __thread_join(el_val_t tid_v) { - int slot = (int)(int64_t)tid_v; - if (slot < 0 || slot >= EL_THREAD_MAX) return EL_STR(""); - pthread_join(_threads[slot].tid, NULL); - el_val_t result = _threads[slot].arg->result; - free(_threads[slot].arg); - _threads[slot].alive = 0; - return result; -} - -/* Mutex table */ - -#define EL_MUTEX_MAX 64 - -typedef struct { - pthread_mutex_t mu; - int allocated; -} ElMutexEntry; - -static ElMutexEntry _mutexes[EL_MUTEX_MAX]; -static int _mutex_count = 0; -static pthread_mutex_t _mutex_alloc_mu = PTHREAD_MUTEX_INITIALIZER; - -el_val_t __mutex_new(void) { - pthread_mutex_lock(&_mutex_alloc_mu); - if (_mutex_count >= EL_MUTEX_MAX) { - pthread_mutex_unlock(&_mutex_alloc_mu); - fprintf(stderr, "[__mutex_new] mutex table full\n"); - return EL_INT(-1); - } - int slot = _mutex_count++; - pthread_mutex_init(&_mutexes[slot].mu, NULL); - _mutexes[slot].allocated = 1; - pthread_mutex_unlock(&_mutex_alloc_mu); - return EL_INT(slot); -} - -void __mutex_lock(el_val_t m_v) { - int slot = (int)(int64_t)m_v; - if (slot < 0 || slot >= EL_MUTEX_MAX || !_mutexes[slot].allocated) return; - pthread_mutex_lock(&_mutexes[slot].mu); -} - -void __mutex_unlock(el_val_t m_v) { - int slot = (int)(int64_t)m_v; - if (slot < 0 || slot >= EL_MUTEX_MAX || !_mutexes[slot].allocated) return; - pthread_mutex_unlock(&_mutexes[slot].mu); -} - -/* ── Channels ─────────────────────────────────────────────────────────────── * - * Buffered MPMC channel backed by a mutex + condvar + circular buffer. - * channel_new(capacity) -> Int (handle) - * channel_send(ch, msg) — blocks if full (capacity > 0) or never (unbounded) - * channel_recv(ch) -> String — blocks until a message is available - * channel_try_recv(ch) -> String — non-blocking, returns "" if empty - * channel_close(ch) — signal no more sends; recv drains remaining - * - * Bounded channels (cap > 0): circular buffer, sender blocks when full. - * Unbounded channels (cap == 0): dynamic array, sender never blocks. - */ -#define EL_CHANNEL_MAX 64 -#define EL_CHANNEL_BUF 1024 - -typedef struct { - char** buf; - int cap; /* 0 = unbounded (grows dynamically) */ - int head, tail, count; - int dyn_cap; /* allocated slots for unbounded mode */ - int closed; - pthread_mutex_t mu; - pthread_cond_t not_empty; - pthread_cond_t not_full; -} ElChannel; - -static ElChannel _channels[EL_CHANNEL_MAX]; -static int _channel_count = 0; -static pthread_mutex_t _channel_alloc_mu = PTHREAD_MUTEX_INITIALIZER; - -el_val_t __channel_new(el_val_t capacity_v) { - int cap = (int)(int64_t)capacity_v; - if (cap < 0) cap = 0; - - pthread_mutex_lock(&_channel_alloc_mu); - if (_channel_count >= EL_CHANNEL_MAX) { - pthread_mutex_unlock(&_channel_alloc_mu); - fprintf(stderr, "[__channel_new] channel table full\n"); - return EL_INT(-1); - } - int slot = _channel_count++; - pthread_mutex_unlock(&_channel_alloc_mu); - - ElChannel* ch = &_channels[slot]; - memset(ch, 0, sizeof(*ch)); - ch->cap = cap; - ch->closed = 0; - ch->head = 0; - ch->tail = 0; - ch->count = 0; - - if (cap > 0) { - /* Bounded: fixed circular buffer. */ - ch->buf = (char**)malloc((size_t)cap * sizeof(char*)); - ch->dyn_cap = cap; - } else { - /* Unbounded: start with EL_CHANNEL_BUF slots, grow as needed. */ - ch->buf = (char**)malloc(EL_CHANNEL_BUF * sizeof(char*)); - ch->dyn_cap = EL_CHANNEL_BUF; - } - if (!ch->buf) { - fprintf(stderr, "[__channel_new] out of memory\n"); - return EL_INT(-1); - } - - pthread_mutex_init(&ch->mu, NULL); - pthread_cond_init(&ch->not_empty, NULL); - pthread_cond_init(&ch->not_full, NULL); - - return EL_INT(slot); -} - -void __channel_send(el_val_t ch_v, el_val_t msg_v) { - int slot = (int)(int64_t)ch_v; - if (slot < 0 || slot >= EL_CHANNEL_MAX) return; - ElChannel* ch = &_channels[slot]; - - const char* msg = EL_CSTR(msg_v); - if (!msg) msg = ""; - char* copy = strdup(msg); /* channel owns the string */ - - pthread_mutex_lock(&ch->mu); - - if (ch->closed) { - /* Send on closed channel is a no-op (drop the message). */ - pthread_mutex_unlock(&ch->mu); - free(copy); - return; - } - - if (ch->cap > 0) { - /* Bounded: block while full. */ - while (ch->count >= ch->cap && !ch->closed) { - pthread_cond_wait(&ch->not_full, &ch->mu); - } - if (ch->closed) { - pthread_mutex_unlock(&ch->mu); - free(copy); - return; - } - ch->buf[ch->tail] = copy; - ch->tail = (ch->tail + 1) % ch->cap; - ch->count++; - } else { - /* Unbounded: grow the buffer if needed. */ - if (ch->count >= ch->dyn_cap) { - int new_cap = ch->dyn_cap * 2; - char** grown = (char**)realloc(ch->buf, (size_t)new_cap * sizeof(char*)); - if (!grown) { - pthread_mutex_unlock(&ch->mu); - free(copy); - fprintf(stderr, "[__channel_send] out of memory growing channel\n"); - return; - } - /* The circular buffer may have wrapped. Linearise it first. - * In unbounded mode head is always 0 (we append at tail, drain - * from head), so a simple memmove isn't needed — but if the - * buffer did wrap (tail < head after growth), we need to fix up. - * Simplest safe path: if tail wrapped, move the head..old_cap - * segment to new_cap..new_cap+(old_cap-head). */ - if (ch->tail < ch->head) { - /* Wrapped: [head..old_cap) is the front, [0..tail) is the back. */ - int front = ch->dyn_cap - ch->head; - memmove(grown + ch->dyn_cap, grown + ch->head, (size_t)front * sizeof(char*)); - ch->head = ch->dyn_cap; - } - ch->buf = grown; - ch->dyn_cap = new_cap; - } - ch->buf[ch->tail] = copy; - ch->tail = (ch->tail + 1) % ch->dyn_cap; - ch->count++; - } - - pthread_cond_signal(&ch->not_empty); - pthread_mutex_unlock(&ch->mu); -} - -el_val_t __channel_recv(el_val_t ch_v) { - int slot = (int)(int64_t)ch_v; - if (slot < 0 || slot >= EL_CHANNEL_MAX) return EL_STR(""); - ElChannel* ch = &_channels[slot]; - - pthread_mutex_lock(&ch->mu); - - /* Block until there is a message or the channel is closed and drained. */ - while (ch->count == 0 && !ch->closed) { - pthread_cond_wait(&ch->not_empty, &ch->mu); - } - - if (ch->count == 0) { - /* Closed and empty — signal EOF. */ - pthread_mutex_unlock(&ch->mu); - return EL_STR(""); - } - - int buf_cap = (ch->cap > 0) ? ch->cap : ch->dyn_cap; - char* msg = ch->buf[ch->head]; - ch->head = (ch->head + 1) % buf_cap; - ch->count--; - - pthread_cond_signal(&ch->not_full); - pthread_mutex_unlock(&ch->mu); - - /* Hand the string to the arena so it is freed after the request. */ - el_arena_track(msg); - return EL_STR(msg); -} - -el_val_t __channel_try_recv(el_val_t ch_v) { - int slot = (int)(int64_t)ch_v; - if (slot < 0 || slot >= EL_CHANNEL_MAX) return EL_STR(""); - ElChannel* ch = &_channels[slot]; - - pthread_mutex_lock(&ch->mu); - - if (ch->count == 0) { - pthread_mutex_unlock(&ch->mu); - return EL_STR(""); - } - - int buf_cap = (ch->cap > 0) ? ch->cap : ch->dyn_cap; - char* msg = ch->buf[ch->head]; - ch->head = (ch->head + 1) % buf_cap; - ch->count--; - - pthread_cond_signal(&ch->not_full); - pthread_mutex_unlock(&ch->mu); - - el_arena_track(msg); - return EL_STR(msg); -} - -void __channel_close(el_val_t ch_v) { - int slot = (int)(int64_t)ch_v; - if (slot < 0 || slot >= EL_CHANNEL_MAX) return; - ElChannel* ch = &_channels[slot]; - - pthread_mutex_lock(&ch->mu); - ch->closed = 1; - /* Wake all blocked recvers and senders so they can observe the close. */ - pthread_cond_broadcast(&ch->not_empty); - pthread_cond_broadcast(&ch->not_full); - pthread_mutex_unlock(&ch->mu); -} - diff --git a/el-compiler/runtime/el_runtime.h b/el-compiler/runtime/el_runtime.h deleted file mode 100644 index ae34612..0000000 --- a/el-compiler/runtime/el_runtime.h +++ /dev/null @@ -1,799 +0,0 @@ -/* - * el_runtime.h — El language C runtime header - * - * Declares all built-in functions available to compiled El programs. - * Include this in every generated .c file. - * - * Value model: - * All El values are represented as el_val_t (= int64_t). - * On 64-bit systems a pointer fits in int64_t. - * String values are cast: (el_val_t)(uintptr_t)"hello" - * Integer values are stored directly. - * This lets arithmetic work naturally while still passing strings around. - * - * Type conventions (El -> C): - * String -> el_val_t (holds const char* via uintptr_t cast) - * Int -> el_val_t - * Bool -> el_val_t (0 = false, nonzero = true) - * Any -> el_val_t - * Void -> void - * - * Macros for convenience: - * EL_STR(s) cast string literal to el_val_t - * EL_CSTR(v) cast el_val_t back to const char* - * EL_INT(v) identity — el_val_t is already int64_t - * - * Link requirements: - * -lcurl — required for the HTTP client (http_get, http_post, llm_*). - * -lpthread — required for the HTTP server (one detached thread per - * connection, capped at 64 concurrent). - * -loqs — optional; required only when liboqs is installed and the - * pq_* / sha3_256_hex entry points are needed. Detected at - * compile time via __has_include(). - * -lcrypto — optional; pulled in alongside -loqs. Used for X25519 in - * pq_hybrid_* and HKDF-SHA256 derivation. - * - * Canonical compile command: - * cc -std=c11 -I el-compiler/runtime -lcurl -lpthread \ - * -o .c el-compiler/runtime/el_runtime.c - * - * With liboqs (post-quantum stack): - * cc -std=c11 -I el-compiler/runtime -lcurl -lpthread -loqs -lcrypto \ - * -o .c el-compiler/runtime/el_runtime.c - */ - -#pragma once - -#include -#include - -typedef int64_t el_val_t; - -#define EL_STR(s) ((el_val_t)(uintptr_t)(s)) -#define EL_CSTR(v) ((const char*)(uintptr_t)(v)) -#define EL_INT(v) (v) -#define EL_NULL ((el_val_t)0) - -/* Float values share the el_val_t (int64) slot via a bit-cast. - * The codegen emits Float literals as `el_from_float()` so the - * underlying bits represent the IEEE 754 double. Float-aware builtins - * (math, format, json) round-trip via these helpers. */ -static inline double el_to_float(el_val_t v) { - union { int64_t i; double f; } u; - u.i = (int64_t)v; - return u.f; -} - -static inline el_val_t el_from_float(double f) { - union { double f; int64_t i; } u; - u.f = f; - return (el_val_t)u.i; -} - -#ifdef __cplusplus -extern "C" { -#endif - -/* ── I/O ──────────────────────────────────────────────────────────────────── */ - -void println(el_val_t s); -void print(el_val_t s); -el_val_t readline(void); - -/* __read_n — read exactly n bytes from stdin; returns String of length n. - * Returns "" on EOF or read error. Used by the El LSP server to read - * JSON-RPC message bodies after the Content-Length header is parsed. */ -el_val_t __read_n(el_val_t n); - -/* __print_raw — write a string to stdout using fwrite + fflush, preserving - * embedded \r\n byte pairs as-is. Used by the El LSP server to emit - * Content-Length framed JSON-RPC messages. */ -void __print_raw(el_val_t s); - -/* ── String builtins ─────────────────────────────────────────────────────── */ - -el_val_t el_str_concat(el_val_t a, el_val_t b); -el_val_t str_eq(el_val_t a, el_val_t b); -el_val_t str_starts_with(el_val_t s, el_val_t prefix); -el_val_t str_ends_with(el_val_t s, el_val_t suffix); -el_val_t str_len(el_val_t s); -el_val_t str_concat(el_val_t a, el_val_t b); -el_val_t int_to_str(el_val_t n); -el_val_t str_to_int(el_val_t s); -el_val_t str_slice(el_val_t s, el_val_t start, el_val_t end); -el_val_t str_contains(el_val_t s, el_val_t sub); -el_val_t str_replace(el_val_t s, el_val_t from, el_val_t to); -el_val_t str_to_upper(el_val_t s); -el_val_t str_to_lower(el_val_t s); -el_val_t str_trim(el_val_t s); - -/* ── Math ────────────────────────────────────────────────────────────────── */ - -el_val_t el_abs(el_val_t n); -el_val_t el_max(el_val_t a, el_val_t b); -el_val_t el_min(el_val_t a, el_val_t b); - -/* ── Refcount (ARC) ────────────────────────────────────────────────────────── - * Lists and Maps carry a refcount. Strings and ints do not — el_retain and - * el_release are safe no-ops on non-refcounted values (they sniff a magic - * header at offset 0 and only act if the magic matches). - * - * Codegen emits these at let-binding shadowing, function entry (params), and - * function exit (locals other than the returned value). The refcount lets - * el_list_append and el_map_set mutate in place when uniquely owned (cheap) - * and copy-on-write when shared (preserves persistent semantics across - * accumulator patterns in the compiler itself). */ - -void el_retain(el_val_t v); -void el_release(el_val_t v); - -/* ── List ────────────────────────────────────────────────────────────────── */ - -el_val_t el_list_new(el_val_t count, ...); -el_val_t el_list_len(el_val_t list); -el_val_t el_list_get(el_val_t list, el_val_t index); -el_val_t el_list_append(el_val_t list, el_val_t elem); -el_val_t el_list_empty(void); -el_val_t el_list_clone(el_val_t list); - -/* ── Map ─────────────────────────────────────────────────────────────────── */ - -el_val_t el_map_new(el_val_t pair_count, ...); -el_val_t el_get_field(el_val_t map, el_val_t key); -el_val_t el_map_get(el_val_t map, el_val_t key); -el_val_t el_map_set(el_val_t map, el_val_t key, el_val_t value); - -/* ── HTTP ─────────────────────────────────────────────────────────────────── */ - -el_val_t http_get(el_val_t url); -el_val_t http_post(el_val_t url, el_val_t body); -el_val_t http_post_json(el_val_t url, el_val_t json_body); -el_val_t http_get_with_headers(el_val_t url, el_val_t headers_map); -el_val_t http_post_with_headers(el_val_t url, el_val_t body, el_val_t headers_map); -el_val_t http_post_form_auth(el_val_t url, el_val_t form_body, el_val_t auth_header); -el_val_t http_delete(el_val_t url); -void http_serve(el_val_t port, el_val_t handler); -void http_set_handler(el_val_t name); - -/* HTTP server v2 ───────────────────────────────────────────────────────────── - * Same dispatch model as http_serve, but the handler signature is widened: - * - * el_val_t handler(method, path, headers_map, body) - * - * `headers_map` is an ElMap from lowercased header name → header value (both - * Strings). Repeated headers are joined with ", " per RFC 7230. - * - * Response value: the handler may return either - * (a) a plain body string — same auto-content-type / 200-OK behaviour as - * http_serve (3-arg) — or - * (b) a response envelope built with `http_response(status, headers_json, - * body)`. The runtime detects the envelope discriminator - * `"el_http_response":1` at the start of the returned string and - * unpacks status / headers / body before sending. - * - * The 3-arg http_serve(port, handler) remains supported unchanged for - * existing handlers (e.g. products/web/server.el): it dispatches with - * (method, path, body), hardcodes 200 OK, and auto-detects content type. */ -void http_serve_v2(el_val_t port, el_val_t handler); -void http_set_handler_v2(el_val_t name); - -/* Build an HTTP response envelope. `headers_json` should be a JSON object - * literal like `{"WWW-Authenticate":"Basic"}` (or "" / "{}" for none). The - * returned string carries the discriminator `{"el_http_response":1,...}` - * which the runtime's send-path detects and unpacks. Detection happens - * uniformly inside http_send_response, so a 3-arg handler may also return - * an envelope. The 3-arg variant remains documented as a fixed 200-OK - * auto-content-type contract for legacy handlers that return plain bodies. */ -el_val_t http_response(el_val_t status, el_val_t headers_json, el_val_t body); - -/* HTTP timeout — every libcurl request honors EL_HTTP_TIMEOUT_MS (default - * 60000ms). Read lazily on first use, so setting the env var any time before - * the first http_* call is sufficient. */ - -/* Streaming variants — write the response body straight to a file via - * libcurl's CURLOPT_WRITEFUNCTION = fwrite. These bypass the el_val_t string - * wrapper entirely, so binary payloads (audio/mpeg, image/png, etc.) survive - * embedded NUL bytes that would truncate a strlen()-based code path. - * - * Both honor EL_HTTP_TIMEOUT_MS, follow redirects, and accept the same - * `headers_map` shape as http_post_with_headers (ElMap of String→String). - * - * Return value: 1 on success (file fully written), 0 on any failure - * (network, file open, partial write). On failure the output file is removed - * so callers cannot mistake a partially-written file for a valid one. */ -el_val_t http_post_to_file(el_val_t url, el_val_t body, el_val_t headers_map, el_val_t output_path); -el_val_t http_get_to_file(el_val_t url, el_val_t headers_map, el_val_t output_path); - -/* ── URL encoding ────────────────────────────────────────────────────────── */ - -el_val_t url_encode(el_val_t s); /* RFC 3986 unreserved set */ -el_val_t url_decode(el_val_t s); /* '+' → space, %XX → byte */ - -/* ── HTML allowlist sanitizer ──────────────────────────────────────────────── - * el_html_sanitize(input_html, allowlist_json) — strict allowlist HTML - * cleaner. State-machine parser; tag/attribute names compared case- - * insensitively against the allowlist; `` / `<… src>` URL schemes - * validated (http, https, mailto, fragment-only, or relative); whole- - * subtree drop for script / style / iframe / object / embed / form; HTML- - * escapes free text outside dropped subtrees. - * - * The allowlist is JSON of the form - * {"p":[],"a":["href","title"],"strong":[],...} - * where each value is the array of attribute names allowed for that tag. */ -el_val_t el_html_sanitize(el_val_t input_html, el_val_t allowlist_json); - -/* ── Filesystem ──────────────────────────────────────────────────────────── */ - -el_val_t fs_read(el_val_t path); -el_val_t fs_write(el_val_t path, el_val_t content); -el_val_t fs_list(el_val_t path); -el_val_t fs_exists(el_val_t path); -el_val_t fs_mkdir(el_val_t path); /* mkdir -p, mode 0755 */ - -/* Length-explicit binary write. `length` is an Int (el_val_t holding the - * byte count). The caller knows the length from context — typically because - * `bytes` came from base64_decode (which produces a magic-tagged binary - * buffer with embedded NULs possible) and the caller already tracks the - * decoded length, OR because the bytes came from a fixed-size source - * (sha256_bytes = 32, hmac_sha256_bytes = 32). Bypasses strlen entirely. - * - * Returns 1 on success, 0 on failure (invalid path, can't open, partial - * write, negative length). On partial-write failure, the file is removed - * so callers cannot read back a truncated artefact. */ -el_val_t fs_write_bytes(el_val_t path, el_val_t bytes, el_val_t length); - -/* ── JSON ────────────────────────────────────────────────────────────────── */ - -el_val_t json_get(el_val_t json, el_val_t key); -el_val_t json_parse(el_val_t s); -el_val_t json_stringify(el_val_t v); -el_val_t json_get_string(el_val_t json_str, el_val_t key); -el_val_t json_get_int(el_val_t json_str, el_val_t key); -el_val_t json_get_float(el_val_t json_str, el_val_t key); -el_val_t json_get_bool(el_val_t json_str, el_val_t key); -el_val_t json_get_raw(el_val_t json_str, el_val_t key); -el_val_t json_set(el_val_t json_str, el_val_t key, el_val_t value); -el_val_t json_array_len(el_val_t json_str); -el_val_t json_array_get(el_val_t json_str, el_val_t index); -el_val_t json_array_get_string(el_val_t json_str, el_val_t index); - -/* ── Time ────────────────────────────────────────────────────────────────── */ - -el_val_t time_now(void); -el_val_t time_now_utc(void); -el_val_t sleep_secs(el_val_t secs); -el_val_t sleep_ms(el_val_t ms); -el_val_t time_format(el_val_t ts, el_val_t fmt); -el_val_t time_to_parts(el_val_t ts); -el_val_t time_from_parts(el_val_t secs, el_val_t ns, el_val_t tz); -el_val_t time_add(el_val_t ts, el_val_t n, el_val_t unit); -el_val_t time_diff(el_val_t ts1, el_val_t ts2, el_val_t unit); - -/* ── Instant + Duration: first-class temporal types ────────────────────────── - * Both types share the el_val_t (int64) slot. Instants are nanoseconds - * since the Unix epoch; Durations are signed nanoseconds. Type discipline - * is enforced at codegen-time: BinOps on names registered as Instant or - * Duration route through the typed wrappers below; mismatches like - * Instant+Instant become #error at the C compiler. - * - * Postfix literals — `30.seconds`, `1.hour`, `500.millis`, `30.nanos` — are - * recognised by the parser as DurationLit AST nodes and lowered to literal - * int64 nanoseconds at codegen time. The runtime never sees the units. */ - -el_val_t el_now_instant(void); -el_val_t now(void); -el_val_t unix_seconds(el_val_t n); -el_val_t unix_millis(el_val_t n); -el_val_t instant_from_iso8601(el_val_t s); - -el_val_t el_duration_from_nanos(el_val_t ns); -el_val_t duration_seconds(el_val_t n); -el_val_t duration_millis(el_val_t n); -el_val_t duration_nanos(el_val_t n); - -el_val_t el_instant_add_dur(el_val_t inst, el_val_t dur); -el_val_t el_instant_sub_dur(el_val_t inst, el_val_t dur); -el_val_t el_instant_diff(el_val_t a, el_val_t b); -el_val_t el_duration_add(el_val_t a, el_val_t b); -el_val_t el_duration_sub(el_val_t a, el_val_t b); -el_val_t el_duration_scale(el_val_t dur, el_val_t scalar); -el_val_t el_duration_div(el_val_t dur, el_val_t scalar); - -el_val_t el_instant_lt(el_val_t a, el_val_t b); -el_val_t el_instant_le(el_val_t a, el_val_t b); -el_val_t el_instant_gt(el_val_t a, el_val_t b); -el_val_t el_instant_ge(el_val_t a, el_val_t b); -el_val_t el_instant_eq(el_val_t a, el_val_t b); -el_val_t el_instant_ne(el_val_t a, el_val_t b); -el_val_t el_duration_lt(el_val_t a, el_val_t b); -el_val_t el_duration_le(el_val_t a, el_val_t b); -el_val_t el_duration_gt(el_val_t a, el_val_t b); -el_val_t el_duration_ge(el_val_t a, el_val_t b); -el_val_t el_duration_eq(el_val_t a, el_val_t b); -el_val_t el_duration_ne(el_val_t a, el_val_t b); - -el_val_t instant_to_unix_seconds(el_val_t i); -el_val_t instant_to_unix_millis(el_val_t i); -el_val_t instant_to_iso8601(el_val_t i); -el_val_t duration_to_seconds(el_val_t d); -el_val_t duration_to_millis(el_val_t d); -el_val_t duration_to_nanos(el_val_t d); - -el_val_t el_sleep_duration(el_val_t dur); -el_val_t unix_timestamp(void); - -el_val_t ttl_cache_set(el_val_t key, el_val_t value); -el_val_t ttl_cache_get(el_val_t key, el_val_t max_age); -el_val_t ttl_cache_age(el_val_t key); - -/* ── Calendar + CalendarTime + Rhythm + LocalDate/Time/DateTime ───────────── - * Phase 1.5 of the time system. Calendar is pluggable: EarthCalendar (IANA - * zones, Gregorian, DST) is the user-facing default; MarsCalendar, - * CycleCalendar(period), NoCycleCalendar, RelativeCalendar handle non-Earth - * domains. - * - * A Calendar interprets an Instant under a particular cycle convention and - * produces a CalendarTime. CalendarTime carries the underlying Instant and - * a back-pointer to its Calendar; arithmetic and formatting consult the - * Calendar to convert ns since epoch into year/month/day/hour/minute/second - * (or sol/phase, or cycle/phase, depending on kind). - * - * Storage convention: Calendar / CalendarTime / Rhythm / LocalDate / - * LocalDateTime are heap-allocated structs whose pointers are cast into - * el_val_t. A 24-bit magic header at offset 0 lets the runtime identify - * the kind safely. LocalTime is small enough to live in the int64 slot - * directly (nanos since midnight, signed). */ - -/* Zone — opaque IANA zone or fixed offset, used by EarthCalendar. - * `zone_id` is either an IANA name ("America/New_York", "UTC") or a fixed - * offset string ("+05:30", "-08:00"). The runtime resolves it via tzset() - * on first use of the owning EarthCalendar. */ -el_val_t zone(el_val_t id); -el_val_t zone_utc(void); -el_val_t zone_local(void); -el_val_t zone_offset(el_val_t hours, el_val_t minutes); - -/* Calendar constructors. Each returns an el_val_t pointer to a heap- - * allocated, magic-tagged Calendar struct. Calendars are interned by - * (kind, zone_id, period_ns, epoch_ns) so identical constructors return - * the same pointer — equality is reference equality. */ -el_val_t earth_calendar(el_val_t z); -el_val_t earth_calendar_default(void); -el_val_t mars_calendar(void); -el_val_t cycle_calendar(el_val_t period_dur); -el_val_t no_cycle_calendar(void); -el_val_t relative_calendar(el_val_t epoch_inst); - -/* CalendarTime constructors and methods. Returns a heap-allocated struct - * whose pointer fits in el_val_t. */ -el_val_t now_in(el_val_t cal); -el_val_t in_calendar(el_val_t inst, el_val_t cal); -el_val_t cal_format(el_val_t ct, el_val_t pattern); -el_val_t cal_to_instant(el_val_t ct); -el_val_t cal_cycle_phase(el_val_t ct); -el_val_t cal_in(el_val_t ct, el_val_t cal); - -/* LocalDate / LocalTime / LocalDateTime — calendar-agnostic value types. - * LocalTime carries nanoseconds since midnight as a signed int64 directly - * in the el_val_t slot (no allocation). LocalDate / LocalDateTime are - * heap-allocated structs with magic headers. */ -el_val_t local_date(el_val_t y, el_val_t m, el_val_t d); -el_val_t local_time(el_val_t h, el_val_t m, el_val_t s, el_val_t ns); -el_val_t local_datetime(el_val_t date, el_val_t time); -el_val_t zoned(el_val_t date, el_val_t time, el_val_t cal); - -el_val_t local_date_year(el_val_t ld); -el_val_t local_date_month(el_val_t ld); -el_val_t local_date_day(el_val_t ld); -el_val_t local_time_hour(el_val_t lt); -el_val_t local_time_minute(el_val_t lt); -el_val_t local_time_second(el_val_t lt); -el_val_t local_time_nanos(el_val_t lt); - -el_val_t el_local_date_add_dur(el_val_t ld, el_val_t dur); -el_val_t el_local_time_add_dur(el_val_t lt, el_val_t dur); -el_val_t el_local_date_lt(el_val_t a, el_val_t b); -el_val_t el_local_date_eq(el_val_t a, el_val_t b); - -/* Rhythm — pluggable recurrence AST. Returns a heap-allocated struct - * pointer in el_val_t; rhythms are immutable so callers may share them. */ -el_val_t rhythm_cycle_start(void); -el_val_t rhythm_cycle_phase(el_val_t phase); -el_val_t rhythm_duration(el_val_t d); -el_val_t rhythm_session_start(void); -el_val_t rhythm_event(el_val_t name); -el_val_t rhythm_and(el_val_t a, el_val_t b); -el_val_t rhythm_or(el_val_t a, el_val_t b); -el_val_t rhythm_weekday(el_val_t day); -el_val_t rhythm_weekly_at(el_val_t day, el_val_t hour, el_val_t minute); -el_val_t rhythm_next_after(el_val_t r, el_val_t after, el_val_t cal); -el_val_t rhythm_matches(el_val_t r, el_val_t ct); - -/* ── UUID ────────────────────────────────────────────────────────────────── */ - -el_val_t uuid_new(void); -el_val_t uuid_v4(void); - -/* ── Environment ─────────────────────────────────────────────────────────── */ - -el_val_t env(el_val_t key); - -/* ── In-process state K/V ────────────────────────────────────────────────── */ - -el_val_t state_set(el_val_t key, el_val_t value); -el_val_t state_get(el_val_t key); -el_val_t state_del(el_val_t key); -el_val_t state_keys(void); - -/* ── Float formatting ────────────────────────────────────────────────────── */ - -el_val_t float_to_str(el_val_t f); -el_val_t int_to_float(el_val_t n); -el_val_t float_to_int(el_val_t f); -el_val_t format_float(el_val_t f, el_val_t decimals); -el_val_t decimal_round(el_val_t f, el_val_t decimals); -el_val_t str_to_float(el_val_t s); - -/* ── Math (Float-aware) ──────────────────────────────────────────────────── */ - -el_val_t math_sqrt(el_val_t f); -el_val_t math_log(el_val_t f); -el_val_t math_ln(el_val_t f); -el_val_t math_sin(el_val_t f); -el_val_t math_cos(el_val_t f); -el_val_t math_pi(void); - -/* ── String additions ────────────────────────────────────────────────────── */ - -el_val_t str_index_of(el_val_t s, el_val_t sub); -el_val_t str_split(el_val_t s, el_val_t sep); -el_val_t str_char_at(el_val_t s, el_val_t i); -el_val_t str_char_code(el_val_t s, el_val_t i); -el_val_t str_pad_left(el_val_t s, el_val_t width, el_val_t pad); -el_val_t str_pad_right(el_val_t s, el_val_t width, el_val_t pad); -el_val_t str_format(el_val_t fmt, el_val_t data); -el_val_t str_lower(el_val_t s); -el_val_t str_upper(el_val_t s); - -/* ── Text-processing primitives (Phase 1: byte/codepoint, ASCII char classes) - * Phase 2 (filed): Unicode-grapheme awareness, NFC/NFD normalization, regex. - * is_* predicates: empty input returns false; multi-char requires ALL bytes - * to match. ASCII ranges only in Phase 1. */ - -/* Counting */ -el_val_t str_count(el_val_t s, el_val_t sub); /* non-overlapping */ -el_val_t str_count_chars(el_val_t s); /* codepoint count */ -el_val_t str_count_bytes(el_val_t s); /* alias of str_len */ -el_val_t str_count_lines(el_val_t s); -el_val_t str_count_words(el_val_t s); -el_val_t str_count_letters(el_val_t s); /* ASCII [A-Za-z] */ -el_val_t str_count_digits(el_val_t s); /* ASCII [0-9] */ - -/* Find / position */ -el_val_t str_index_of_all(el_val_t s, el_val_t sub); /* [Int] of byte offsets */ -el_val_t str_last_index_of(el_val_t s, el_val_t sub); -el_val_t str_find_chars(el_val_t s, el_val_t any_of); /* first idx of any ch */ - -/* Transform */ -el_val_t str_repeat(el_val_t s, el_val_t n); -el_val_t str_reverse(el_val_t s); /* by codepoint */ -el_val_t str_strip_prefix(el_val_t s, el_val_t prefix); -el_val_t str_strip_suffix(el_val_t s, el_val_t suffix); -el_val_t str_strip_chars(el_val_t s, el_val_t chars); -el_val_t str_lstrip(el_val_t s); -el_val_t str_rstrip(el_val_t s); - -/* Char classification (Bool) */ -el_val_t is_letter(el_val_t s); -el_val_t is_digit(el_val_t s); -el_val_t is_alphanumeric(el_val_t s); -el_val_t is_whitespace(el_val_t s); -el_val_t is_punctuation(el_val_t s); -el_val_t is_uppercase(el_val_t s); -el_val_t is_lowercase(el_val_t s); - -/* Split / join */ -el_val_t str_split_lines(el_val_t s); -el_val_t str_split_chars(el_val_t s); /* alias of native_string_chars */ -el_val_t str_split_n(el_val_t s, el_val_t sep, el_val_t n); -el_val_t str_join(el_val_t list, el_val_t sep); /* alias of list_join */ - -/* ── List additions ──────────────────────────────────────────────────────── */ - -el_val_t list_push(el_val_t list, el_val_t elem); -el_val_t list_push_front(el_val_t list, el_val_t elem); -el_val_t list_join(el_val_t list, el_val_t sep); -el_val_t list_range(el_val_t start, el_val_t end); - -/* ── Bool helpers ────────────────────────────────────────────────────────── */ - -el_val_t bool_to_str(el_val_t b); - -/* ── Numeric parsing ─────────────────────────────────────────────────────── */ - -el_val_t parse_int(el_val_t s, el_val_t default_val); - -/* ── Process ─────────────────────────────────────────────────────────────── */ - -void exit_program(el_val_t code); -el_val_t getpid_now(void); - -/* ── CGI identity ───────────────────────────────────────────────────────────── - * Called at the start of main() in CGI programs (those with a `cgi {}` block). - * Records the program's DHARMA identity before any other code executes. */ - -void el_cgi_init(el_val_t name, el_val_t dharma_id, el_val_t principal, - el_val_t network, el_val_t engram); - -/* ── DHARMA network builtins ───────────────────────────────────────────────── - * Available to CGI programs (declared with a `cgi {}` block). - * - * Peers are addressed by `dharma_id` of the form - * "@" e.g. "ntn-genesis@http://localhost:7770" - * If the @ portion is omitted, transport defaults to - * "http://localhost:7770" (the local CGI daemon assumption). - * - * Wire protocol (all peers expose): - * POST /dharma/recv { channel, from, content } → response body - * POST /dharma/event { type, payload, source, timestamp } - * POST /api/activate { query } → list of nodes - * - * Hosting application's responsibility: an El program with a `cgi {}` block - * runs http_serve() with its own request handler; that handler should route - * "/dharma/event" requests by calling el_runtime_dharma_event_arrive() so - * incoming events feed dharma_field() queues. The runtime itself does not - * intercept any /dharma path. */ - -el_val_t dharma_connect(el_val_t cgi_id); -el_val_t dharma_send(el_val_t channel, el_val_t content); -el_val_t dharma_activate(el_val_t query); -void dharma_emit(el_val_t event_type, el_val_t payload); -el_val_t dharma_field(el_val_t event_type); -void dharma_strengthen(el_val_t cgi_id, el_val_t weight); -el_val_t dharma_relationship(el_val_t cgi_id); -el_val_t dharma_peers(void); - -/* Public C API: called by an El program's HTTP handler when a /dharma/event - * request arrives. Pushes onto the per-event-type queue and signals any - * pending dharma_field() blockers. All three arguments must be NUL-terminated - * C strings (or NULL — then treated as empty). */ -void el_runtime_dharma_event_arrive(const char* event_type, - const char* payload, - const char* source); - -/* ── Engram local graph primitives ─────────────────────────────────────────── - * Operate on the CGI's local Engram knowledge graph. - * `engram_activate` queries the local graph only; `dharma_activate` is - * network-wide across all connected CGI graphs. */ - -el_val_t engram_node(el_val_t content, el_val_t node_type, el_val_t salience); -el_val_t engram_node_full(el_val_t content, el_val_t node_type, el_val_t label, - el_val_t salience, el_val_t importance, el_val_t confidence, - el_val_t tier, el_val_t tags); -/* Layered consciousness — see el_runtime.c for the layered architecture - * design notes (search "Layered consciousness architecture"). The five - * canonical layers (safety / core-identity / domain-knowledge / imprint / - * suit) are seeded automatically; engram_add_layer extends the registry - * with imprint or suit overlays at runtime. Nodes default to layer 1 - * (core-identity) when created via engram_node / engram_node_full. */ -el_val_t engram_node_layered(el_val_t content, el_val_t node_type, el_val_t label, - el_val_t salience, el_val_t certainty, el_val_t confidence, - el_val_t status, el_val_t tags, el_val_t layer_id); -el_val_t engram_add_layer(el_val_t name, el_val_t priority, el_val_t suppressible, - el_val_t transparent, el_val_t injectable); -el_val_t engram_remove_layer(el_val_t layer_id); -el_val_t engram_list_layers(void); -el_val_t engram_get_node(el_val_t id); -void engram_strengthen(el_val_t node_id); -void engram_forget(el_val_t node_id); -el_val_t engram_node_count(void); -el_val_t engram_search(el_val_t query, el_val_t limit); -el_val_t engram_scan_nodes(el_val_t limit, el_val_t offset); -void engram_connect(el_val_t from_id, el_val_t to_id, el_val_t weight, el_val_t relation); -el_val_t engram_edge_between(el_val_t from_id, el_val_t to_id); -el_val_t engram_neighbors(el_val_t node_id); -el_val_t engram_neighbors_filtered(el_val_t node_id, el_val_t max_depth, el_val_t direction); -el_val_t engram_edge_count(void); -/* Three-pass activation: background fan-out → working-memory promotion → - * Layer 0 override. See "Three-pass activation" in el_runtime.c. */ -el_val_t engram_activate(el_val_t query, el_val_t depth); -el_val_t engram_save(el_val_t path); -el_val_t engram_load(el_val_t path); - -/* JSON-string accessors — return pre-serialized JSON so HTTP handlers - * can pass results straight through without round-tripping ElList/ElMap - * through json_stringify. */ -el_val_t engram_get_node_json(el_val_t id); -el_val_t engram_search_json(el_val_t query, el_val_t limit); -el_val_t engram_scan_nodes_json(el_val_t limit, el_val_t offset); -el_val_t engram_scan_nodes_by_type_json(el_val_t node_type, el_val_t limit, el_val_t offset); -el_val_t engram_neighbors_json(el_val_t node_id, el_val_t max_depth, el_val_t direction); -el_val_t engram_activate_json(el_val_t query, el_val_t depth); -el_val_t engram_stats_json(void); -el_val_t engram_list_layers_json(void); -/* engram_compile_layered_json — produce a prompt-ready text block split - * into "[LAYER 0 — STRUCTURAL]" (non-suppressible layers, sacred fire) - * and "[ENGRAM CONTEXT]" (standard suppressible layers). Returns "" if - * no nodes promoted to working memory. */ -el_val_t engram_compile_layered_json(el_val_t intent, el_val_t depth); - -/* ── LLM (Anthropic API client) ───────────────────────────────────────────── - * All functions call https://api.anthropic.com/v1/messages with the API key - * from env ANTHROPIC_API_KEY. Default model when empty: claude-sonnet-4-5. */ - -el_val_t llm_call(el_val_t model, el_val_t prompt); -el_val_t llm_call_system(el_val_t model, el_val_t system_prompt, el_val_t user_prompt); -el_val_t llm_call_agentic(el_val_t model, el_val_t system, el_val_t user, el_val_t tools); -el_val_t llm_vision(el_val_t model, el_val_t system, el_val_t prompt, el_val_t image_url_or_b64); -el_val_t llm_models(void); - -/* Register a tool handler by name. The handler is looked up via dlsym - * (mirroring http_set_handler), so any El `fn (input)` compiles to - * a global C symbol that this function can locate at runtime. - * Handler signature: `el_val_t handler(el_val_t input_json)` — receives - * the tool input as a JSON-string el_val_t and returns a JSON-string - * el_val_t result. Used by llm_call_agentic. */ -void llm_register_tool(el_val_t name, el_val_t handler_fn_name); - -/* ── args() ───────────────────────────────────────────────────────────────── - * Provides access to command-line arguments passed to the program. - * Populated by el_runtime_init_args() before main() runs. */ - -el_val_t args(void); -void el_runtime_init_args(int argc, char** argv); - -/* ── Crypto primitives ───────────────────────────────────────────────────── - * SHA-256, HMAC-SHA-256, and base64 (standard + URL-safe). - * Self-contained — no OpenSSL/libcrypto dependency. The implementations are - * adapted from public-domain reference code (Brad Conte / RFC 4648). - * - * Bytes-returning variants (sha256_bytes, hmac_sha256_bytes) return a string - * value whose contents are raw binary; callers usually feed these into - * base64_encode. Note that el_val_t strings are NUL-terminated by convention, - * so the binary payload may contain embedded NULs — pass it directly into - * base64_encode (which uses an explicit length) rather than treating it as - * a printable C string. - * - * The "base64" variants emit/accept RFC 4648 standard alphabet with padding. - * The "base64url" variants use URL-safe alphabet (`-`/`_`) with no padding, - * as used in JWTs. */ - -el_val_t sha256_hex(el_val_t input); -el_val_t sha256_bytes(el_val_t input); -el_val_t hmac_sha256_hex(el_val_t key, el_val_t message); -el_val_t hmac_sha256_bytes(el_val_t key, el_val_t message); -el_val_t base64_encode(el_val_t input); -el_val_t base64_decode(el_val_t input); -el_val_t base64url_encode(el_val_t input); -el_val_t base64url_decode(el_val_t input); - -/* Length-aware variants (internal — exposed for the rare caller that already - * has a known-length binary buffer and doesn't want to round-trip through - * a NUL-terminated el_val_t string). Sha256_bytes and hmac_sha256_bytes feed - * these implicitly. */ -el_val_t el_sha256_bytes_n(const unsigned char* data, size_t len); -el_val_t el_base64_encode_n(const unsigned char* data, size_t len, int url_safe); - -/* ── Post-quantum primitives (liboqs-backed) ──────────────────────────────── - * All inputs/outputs hex-encoded. Algorithm choices: - * Signature: CRYSTALS-Dilithium-3 (NIST level 3, balanced) - * KEM: CRYSTALS-Kyber-768 (NIST level 3) - * Hash: SHA3-256 (Keccak) (PQ-aware protocols favour SHA3 over SHA2) - * - * If liboqs is not linked (detected via __has_include() at compile - * time), the pq_* entry points return a JSON-shaped error string so callers - * fail loudly rather than silently fall back to classical schemes: - * {"error":"liboqs not linked, post-quantum primitives unavailable"} - * - * The hybrid handshake pairs X25519 with Kyber-768 per NIST PQ guidance and - * CNSA 2.0. Combined shared secret is HKDF-SHA256(x25519_ss || kyber_ss). - * Even if Kyber falls, X25519 holds; if X25519 falls under quantum attack, - * Kyber holds. SHA3-256 also remains usable independent of liboqs (the - * Keccak permutation is PQ-OK as a primitive). */ - -el_val_t pq_keygen_signature(void); -el_val_t pq_sign(el_val_t secret_key_hex, el_val_t message); -el_val_t pq_verify(el_val_t public_key_hex, el_val_t message, el_val_t signature_hex); - -el_val_t pq_kem_keygen(void); -el_val_t pq_kem_encaps(el_val_t public_key_hex); -el_val_t pq_kem_decaps(el_val_t secret_key_hex, el_val_t ciphertext_hex); - -el_val_t pq_hybrid_keygen(void); -el_val_t pq_hybrid_handshake(el_val_t remote_pub_combined); - -el_val_t sha3_256_hex(el_val_t input); - -/* ── AEAD: AES-256-GCM (libcrypto-backed) ─────────────────────────────────── - * Symmetric authenticated encryption used to wrap envelopes after a KEM - * handshake. Caller MUST supply a 32-byte key (64 hex chars) — typically the - * Kyber-768 / hybrid shared_secret, optionally normalized via SHA3-256. - * - * aead_encrypt returns a JSON map {"nonce":"...","ciphertext":"..."} where - * ciphertext is the AES-256-GCM output with the 16-byte auth tag appended. - * Nonce is a fresh 12-byte CSPRNG draw — callers never pick the nonce, which - * structurally rules out the GCM nonce-reuse footgun. - * - * aead_decrypt returns the plaintext String, or "" on any failure (including - * auth-tag mismatch). Callers MUST check for "" before trusting the result. */ -el_val_t aead_encrypt(el_val_t key_hex, el_val_t plaintext); -el_val_t aead_decrypt(el_val_t key_hex, el_val_t nonce_hex, el_val_t ciphertext_hex); - -/* ── Native VM builtin aliases (for compiled El source) ───────────────────── - * These match the El VM's native_* builtins so that El source compiled - * to C can call the same names without modification. */ - -el_val_t native_list_get(el_val_t list, el_val_t index); -el_val_t native_list_len(el_val_t list); -el_val_t native_list_append(el_val_t list, el_val_t elem); -el_val_t native_list_empty(void); -el_val_t native_list_clone(el_val_t list); -el_val_t native_string_chars(el_val_t s); -el_val_t native_int_to_str(el_val_t n); - -/* ── Method-call shorthand aliases ────────────────────────────────────────── - * The El method-call convention `obj.method(args)` compiles to - * `method(obj, args)`. These aliases expose the runtime functions under - * the short names that result from method calls in El source. - * - * Example: `myList.append(x)` → `append(myList, x)` (calls this alias) - * `myList.len()` → `len(myList)` (calls this alias) */ - -el_val_t append(el_val_t list, el_val_t elem); /* el_list_append */ -el_val_t len(el_val_t list); /* el_list_len */ -el_val_t get(el_val_t list, el_val_t index); /* el_list_get */ -el_val_t map_get(el_val_t map, el_val_t key); /* el_map_get */ -el_val_t map_set(el_val_t map, el_val_t key, el_val_t value); /* el_map_set */ - -/* ── OTLP/HTTP Observability ─────────────────────────────────────────────── */ -/* See bottom of el_runtime.c for the implementation. - * Configured by env vars OTLP_ENDPOINT, OTEL_SERVICE_NAME, OTEL_SERVICE_VERSION. - * No-op when OTLP_ENDPOINT is unset. Drop-on-failure semantics. */ -/* ── Subprocess execution ────────────────────────────────────────────────── */ -el_val_t exec_command(el_val_t cmd); /* run shell command, return exit code */ -el_val_t exec_capture(el_val_t cmd); /* run shell command, capture stdout */ -el_val_t exec(el_val_t cmd); /* exec(cmd) → stdout String (30s timeout) */ -el_val_t exec_bg(el_val_t cmd); /* exec_bg(cmd) → PID String (non-blocking) */ - -el_val_t emit_log(el_val_t level, el_val_t msg, el_val_t fields_json); -el_val_t emit_metric(el_val_t name, el_val_t value, el_val_t tags_json); -el_val_t trace_span_start(el_val_t name); -el_val_t trace_span_end(el_val_t span_handle); -el_val_t emit_event(el_val_t name, el_val_t duration_ms); - -/* ── Threading seed primitives ──────────────────────────────────────────────── - * These are the low-level C primitives that back thread.el and channel.el. - * El programs call them via their El wrappers (spawn, join, __mutex_new, etc.) - * rather than directly. - * - * __thread_create(fn_name, arg) — dlsym-resolves fn_name, spawns pthread, - * returns a slot index (Int) usable with __thread_join. - * __thread_join(tid) — joins the thread, returns its String result. - * __mutex_new() — allocates a mutex, returns handle (Int). - * __mutex_lock(m) — locks mutex m (blocks until available). - * __mutex_unlock(m) — unlocks mutex m. */ - -el_val_t __thread_create(el_val_t fn_name, el_val_t arg); -el_val_t __thread_join(el_val_t tid); -el_val_t __mutex_new(void); -void __mutex_lock(el_val_t m); -void __mutex_unlock(el_val_t m); - -/* ── Channel seed primitives ───────────────────────────────────────────────── - * Buffered MPMC channels. All values are Strings; handles are Ints. - * - * __channel_new(capacity) — create channel; cap=0 means unbounded. - * __channel_send(ch, msg) — push msg; blocks if bounded and full. - * __channel_recv(ch) — pop msg; blocks until available; "" on closed+empty. - * __channel_try_recv(ch) — non-blocking pop; "" if empty. - * __channel_close(ch) — mark closed; wakes all blocked recvers/senders. */ - -el_val_t __channel_new(el_val_t capacity); -void __channel_send(el_val_t ch, el_val_t msg); -el_val_t __channel_recv(el_val_t ch); -el_val_t __channel_try_recv(el_val_t ch); -void __channel_close(el_val_t ch); - -#ifdef __cplusplus -} -#endif diff --git a/tools/lsp/build.sh b/tools/lsp/build.sh deleted file mode 100755 index 14b48c4..0000000 --- a/tools/lsp/build.sh +++ /dev/null @@ -1,125 +0,0 @@ -#!/usr/bin/env bash -# tools/lsp/build.sh — Build the El LSP server binary end-to-end. -# -# Pipeline: -# 1. Compile el-lsp.el → C source (using elc) -# 2. Compile C source → binary (using cc with el_runtime.c) -# -# Output: tools/lsp/dist/el-lsp -# -# Prerequisites: -# - dist/platform/elc (the El self-hosted compiler) -# - el-compiler/runtime/el_runtime.c + el_runtime.h -# - cc (clang or gcc), libcurl, pthreads -# -# Usage: -# cd -# ./tools/lsp/build.sh -# -# Or from anywhere: -# EL_HOME=/path/to/el-root ./tools/lsp/build.sh - -set -euo pipefail - -# ── Locate el root ───────────────────────────────────────────────────────── -SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)" -EL_HOME="${EL_HOME:-$(cd "${SCRIPT_DIR}/../.." && pwd)}" - -ELC="${ELC:-${EL_HOME}/dist/platform/elc}" -RUNTIME_DIR="${EL_HOME}/el-compiler/runtime" -LSP_DIR="${SCRIPT_DIR}" -OUT_DIR="${LSP_DIR}/dist" - -# ── Validate prerequisites ───────────────────────────────────────────────── -echo "==> Checking prerequisites..." -echo " EL_HOME = ${EL_HOME}" -echo " elc = ${ELC}" -echo " runtime = ${RUNTIME_DIR}" - -if [ ! -x "${ELC}" ]; then - echo "error: elc not found at ${ELC}" >&2 - echo " Build it first: cd && make (or compile elc-bootstrap.c)" >&2 - exit 1 -fi - -if [ ! -f "${RUNTIME_DIR}/el_runtime.c" ]; then - echo "error: el_runtime.c not found at ${RUNTIME_DIR}" >&2 - exit 1 -fi - -if [ ! -f "${LSP_DIR}/el-lsp.el" ]; then - echo "error: el-lsp.el not found at ${LSP_DIR}" >&2 - exit 1 -fi - -mkdir -p "${OUT_DIR}" - -# ── Compile El → C ───────────────────────────────────────────────────────── -C_OUT="${OUT_DIR}/el-lsp.c" -echo "==> Compiling El → C..." -echo " ${LSP_DIR}/el-lsp.el → ${C_OUT}" - -# el-lsp.el uses only standard El builtins (no import statements needed). -# The elc compiler emits #include "el_runtime.h" at the top of the output. -"${ELC}" "${LSP_DIR}/el-lsp.el" > "${C_OUT}" - -echo " Done ($(wc -l < "${C_OUT}") lines of C)." - -# ── Compile C → binary ───────────────────────────────────────────────────── -BIN="${OUT_DIR}/el-lsp" -echo "==> Compiling C → binary..." -echo " ${C_OUT} + el_runtime.c → ${BIN}" - -cc -std=c11 -O2 \ - -I "${RUNTIME_DIR}" \ - -o "${BIN}" \ - "${C_OUT}" "${RUNTIME_DIR}/el_runtime.c" \ - -lcurl -lpthread - -echo " Done." - -# ── Summary ──────────────────────────────────────────────────────────────── -echo -echo "==> Build complete." -echo -echo " Binary : ${BIN}" -echo " Size : $(du -sh "${BIN}" | cut -f1)" -echo -echo " Install system-wide:" -echo " sudo cp ${BIN} /usr/local/bin/el-lsp" -echo -echo " Quick smoke test (initialize + shutdown):" -cat << 'SMOKETEST' - python3 - << 'PY' -import subprocess, json - -def frame(body): - b = body.encode() - return f"Content-Length: {len(b)}\r\n\r\n".encode() + b - -def send(proc, obj): - proc.stdin.write(frame(json.dumps(obj))) - proc.stdin.flush() - -def recv(proc): - hdr = b"" - while not hdr.endswith(b"\r\n\r\n"): - hdr += proc.stdout.read(1) - cl = int([l for l in hdr.decode().split("\r\n") if l.startswith("Content-Length")][0].split(": ")[1]) - return json.loads(proc.stdout.read(cl)) - -import sys, os -bin_path = sys.argv[1] if len(sys.argv) > 1 else "./dist/el-lsp" -proc = subprocess.Popen([bin_path], stdin=subprocess.PIPE, stdout=subprocess.PIPE) -send(proc, {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"capabilities":{}}}) -r = recv(proc) -print("initialize:", r.get("result", {}).get("serverInfo", {}).get("name"), "OK" if "result" in r else "FAIL") -send(proc, {"jsonrpc":"2.0","id":2,"method":"shutdown","params":{}}) -r = recv(proc) -print("shutdown:", "OK" if r.get("result") is None else "FAIL") -send(proc, {"jsonrpc":"2.0","method":"exit","params":{}}) -proc.wait() -print("exit: OK") -PY -SMOKETEST -