Give cross-cutting concerns an owner instead of a convention (#145)

engram/src/server.el

@@ -10,10 +10,60 @@
 //   cc -std=c11 -O2 -lcurl -lpthread -o engram server.c el_runtime.c
 //   ./engram
 //
-// Configuration via environment:
-//   ENGRAM_BIND       — host:port (default :8742)
-//   ENGRAM_API_KEY    — bearer auth (optional)
-//   ENGRAM_DATA_DIR   — snapshot location (default ~/.neuron/engram)
+// Configuration is DECLARED, not scattered. See the `program` block below:
+// every knob's type and default lives there and nowhere else, is resolved from
+// the environment (env wins, declaration is the fallback) and validated before
+// any statement of this file runs. Read one with config("NAME") -> String.
+//
+// The one deliberate exception is ENGRAM_DATA_DIR — see the note in the block.
+
+// ── Program declaration (cross-cutting concerns) ──────────────────────────────
+//
+// singleton: two engram processes against one data dir is data loss, not a
+// warning. The runtime takes an exclusive flock at startup and a second start
+// is refused loudly with the holder's pid.
+//
+// NOT declared here, on purpose: ENGRAM_DATA_DIR. Its resolution is owned by
+// engram_resolve_data_dir() (el_runtime.c), which defaults to $HOME/.neuron/engram
+// and fails LOUD rather than silently persisting to an ephemeral directory.
+// Declaring a default for it here as well would put the data dir's fallback in
+// two places — which is precisely the defect this migration removes (until
+// 2026-08-15 the reseed backup path carried its own "/tmp/engram" default that
+// disagreed with the resolver, so the pre-destructive safety copy landed in /tmp).
+// HOME is likewise not declared: it is a genuine environment read, not a knob.
+program "engram" {
+    singleton: "engram"
+
+    // ── Core server ──
+    env ENGRAM_BIND: String = ":8742"
+    // Default "" leaves auth DISABLED (check_auth_ok short-circuits to true on an
+    // empty key). That is the pre-existing behaviour and is deliberately preserved
+    // here; making this `required` is the obvious hardening follow-up, but it is a
+    // behaviour change and out of scope for this migration.
+    env ENGRAM_API_KEY: String = ""
+
+    // ── Feature flags (bool-ish Strings; the predicate fns below own truthiness) ──
+    env ENGRAM_STORE: String = "off"
+    env ENGRAM_WAL: String = "off"
+    env ENGRAM_AUTOCONNECT: String = "off"
+    env ENGRAM_ISE_OFFGRAPH: String = "off"
+
+    // ── ISE telemetry ──
+    env ENGRAM_ISE_RETENTION_MS: Int = "172800000"
+
+    // ── Guide (local Qwen3 via llama-server) ──
+    env GUIDE_ENABLE: String = "off"
+    env GUIDE_TIER_FORCE: String = ""
+    env GUIDE_CACHE_DIR: String = ""
+    env GUIDE_RAM_GB_4B: Int = "16"
+    env GUIDE_RAM_GB_1P7B: Int = "8"
+    env GUIDE_BACKEND: String = "llama-server"
+    env GUIDE_HOST: String = "127.0.0.1"
+    env GUIDE_PORT: Int = "8771"
+    env GUIDE_LLAMA_SERVER_BIN: String = "llama-server"
+    env GUIDE_NGL: Int = "99"
+    env GUIDE_CTX: Int = "4096"
+}
 
 // ── Helpers ───────────────────────────────────────────────────────────────────
 
@@ -133,7 +183,7 @@ fn route_text_health(method: String, path: String, body: String) -> String {
 // engram_store_enabled() in el_runtime.c EXACTLY (1 / on / true). Default off →
 // every persistence path below is byte-for-byte the historical snapshot behavior.
 fn store_on() -> Bool {
-    let v: String = env("ENGRAM_STORE")
+    let v: String = config("ENGRAM_STORE")
     if str_eq(v, "1") { return true }
     if str_eq(v, "on") { return true }
     if str_eq(v, "true") { return true }
@@ -162,7 +212,6 @@ fn persist_canonical() -> Int {
     if store_on() {
         return engram_store_checkpoint()
     }
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
     let dir: String = engram_resolve_data_dir()
     // (2026-08-10 self-review) This returned a hardcoded 1, which made every
     // caller's `let saved: Int = persist_canonical()` a dead variable — six
@@ -176,7 +225,7 @@ fn persist_canonical() -> Int {
 // per-write full-snapshot behavior. When ON, structural mutations append O(1)
 // WAL records instead of rewriting the whole graph, with threshold compaction.
 fn wal_on() -> Bool {
-    str_eq(env("ENGRAM_WAL"), "on")
+    str_eq(config("ENGRAM_WAL"), "on")
 }
 
 // autoconnect_on — ENGRAM_AUTOCONNECT. Will's rule: "we shouldn't be inserting
@@ -184,7 +233,7 @@ fn wal_on() -> Bool {
 // edge (kNN over embeddings) so no content node enters the graph edgeless.
 // Default OFF -> byte-identical to prior behavior (node created, no auto edges).
 fn autoconnect_on() -> Bool {
-    let v: String = env("ENGRAM_AUTOCONNECT")
+    let v: String = config("ENGRAM_AUTOCONNECT")
     if str_eq(v, "1") { return true }
     if str_eq(v, "on") { return true }
     if str_eq(v, "true") { return true }
@@ -197,7 +246,7 @@ fn autoconnect_on() -> Bool {
 // separate state-event log tier instead of the node graph. Default OFF -> ISEs
 // remain graph nodes exactly as before (with 48h prune).
 fn ise_offgraph_on() -> Bool {
-    let v: String = env("ENGRAM_ISE_OFFGRAPH")
+    let v: String = config("ENGRAM_ISE_OFFGRAPH")
     if str_eq(v, "1") { return true }
     if str_eq(v, "on") { return true }
     if str_eq(v, "true") { return true }
@@ -394,7 +443,6 @@ fn route_scan_nodes(method: String, path: String, body: String) -> String {
 // process ever booted with a partial/empty store, the first read request
 // clobbered the good snapshot. Read routes must never write the canonical path.)
 fn route_scan_edges(method: String, path: String, body: String) -> String {
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
     let dir: String = engram_resolve_data_dir()
     let snap_path: String = dir + "/.scan-export.json"
     engram_save(snap_path)
@@ -555,7 +603,6 @@ fn route_forget(method: String, path: String, body: String) -> String {
 
 fn route_save(method: String, path: String, body: String) -> String {
     let p_raw: String = json_get_string(body, "path")
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
     let dir: String = engram_resolve_data_dir()
     let p: String = if str_eq(p_raw, "") { dir + "/snapshot.json" } else { p_raw }
     // (2026-08-10 self-review) engram_save returns 0 on an empty path and the
@@ -639,7 +686,6 @@ fn route_drift(method: String, path: String, body: String) -> String {
 
 fn route_load(method: String, path: String, body: String) -> String {
     let p_raw: String = json_get_string(body, "path")
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
     let dir: String = engram_resolve_data_dir()
     let p: String = if str_eq(p_raw, "") { dir + "/snapshot.json" } else { p_raw }
     // (2026-08-10 self-review) This was a stub response over the single most
@@ -710,7 +756,6 @@ fn route_embed_backfill(method: String, path: String, body: String) -> String {
 // (it skips nodes already present by ID). Auth-exempt: same-host internal call.
 // (2026-06-27 self-review: added this route to fix silent 10-min sync failures)
 fn route_sync(method: String, path: String, body: String) -> String {
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
     let dir: String = engram_resolve_data_dir()
     // 2026-07-21 self-review: export to a scratch path, never the canonical
     // snapshot.json — read routes must not be able to clobber the good snapshot.
@@ -786,8 +831,12 @@ fn route_reseed_nodes(method: String, path: String, body: String) -> String {
     if str_eq(p, "") { return err_json("path is required") }
     if str_eq(fs_read(p), "") { return err_json("file missing or empty") }
 
-    let dir_raw: String = env("ENGRAM_DATA_DIR")
-    let dir: String = if str_eq(dir_raw, "") { "/tmp/engram" } else { dir_raw }
+    // (2026-08-15) This site carried its own "/tmp/engram" fallback, which
+    // DISAGREED with engram_resolve_data_dir() ($HOME/.neuron/engram, fail-loud).
+    // The consumer is the pre-destructive backup below, so with ENGRAM_DATA_DIR
+    // unset the safety copy taken before a reseed landed in an ephemeral /tmp
+    // while the store it was protecting lived elsewhere. One owner, one answer.
+    let dir: String = engram_resolve_data_dir()
     let backup: String = dir + "/.reseed-backup.json"
 
     let replace_raw: String = json_get_raw(body, "replace")
@@ -879,8 +928,7 @@ fn route_emit_ise(method: String, path: String, body: String) -> String {
         sal, imp, conf,
         "Episodic", "[\"internal-state\",\"InternalStateEvent\"]"
     )
-    let ret_raw: String = env("ENGRAM_ISE_RETENTION_MS")
-    let ret_ms: Int = if str_eq(ret_raw, "") { 172800000 } else { str_to_int(ret_raw) }
+    let ret_ms: Int = str_to_int(config("ENGRAM_ISE_RETENTION_MS"))
     let pruned: Int = engram_prune_telemetry(ret_ms)
     "{\"ok\":true,\"id\":\"" + id + "\",\"pruned\":" + int_to_str(pruned) + "}"
 }
@@ -1129,14 +1177,12 @@ fn route_correspondence_beat(method: String, path: String, body: String) -> Stri
 // turns native thinking ON: the response carries reasoning_content (the thinking)
 // alongside content (the answer).
 
-fn guide_env_or(key: String, dflt: String) -> String {
-    let v: String = env(key)
-    if str_eq(v, "") { return dflt }
-    return v
-}
+// (2026-08-15) guide_env_or(key, dflt) lived here. Its whole job was supplying a
+// per-call-site default, which is now the program block's job — every GUIDE_* knob
+// is declared once at the top of this file and read straight through config().
 
 fn guide_enabled() -> Bool {
-    let v: String = env("GUIDE_ENABLE")
+    let v: String = config("GUIDE_ENABLE")
     if str_eq(v, "1") { return true }
     if str_eq(v, "on") { return true }
     if str_eq(v, "true") { return true }
@@ -1181,15 +1227,15 @@ fn guide_probe_metal() -> Bool {
 
 // ── 2. Tier selection (config-driven thresholds, spec-autoselected) ────────────
 fn guide_threshold_4b() -> Int {
-    return str_to_int(guide_env_or("GUIDE_RAM_GB_4B", "16"))
+    return str_to_int(config("GUIDE_RAM_GB_4B"))
 }
 fn guide_threshold_1p7b() -> Int {
-    return str_to_int(guide_env_or("GUIDE_RAM_GB_1P7B", "8"))
+    return str_to_int(config("GUIDE_RAM_GB_1P7B"))
 }
 
 // GUIDE_TIER_FORCE overrides the spec autoselect (used to prove cheaply on 0.6b).
 fn guide_select_tier(ram_gb: Int) -> String {
-    let forced: String = env("GUIDE_TIER_FORCE")
+    let forced: String = config("GUIDE_TIER_FORCE")
     if !str_eq(forced, "") { return forced }
     if ram_gb >= guide_threshold_4b() { return "4b" }
     if ram_gb >= guide_threshold_1p7b() { return "1.7b" }
@@ -1209,8 +1255,10 @@ fn guide_file(tier: String) -> String {
 }
 
 fn guide_cache_dir() -> String {
-    let c: String = env("GUIDE_CACHE_DIR")
+    let c: String = config("GUIDE_CACHE_DIR")
     if !str_eq(c, "") { return c }
+    // HOME stays a raw env() read: it is the ambient environment, not a knob of
+    // this program, and it is deliberately absent from the program block.
     let home: String = env("HOME")
     if !str_eq(home, "") { return home + "/.neuron/guide/models" }
     return engram_resolve_data_dir() + "/guide-models"
@@ -1251,9 +1299,9 @@ fn guide_fetch(tier: String) -> Bool {
 }
 
 // ── 4/5. Backend abstraction + BIND as an engageable interlocutor ──────────────
-fn guide_backend() -> String { return guide_env_or("GUIDE_BACKEND", "llama-server") }
-fn guide_host() -> String    { return guide_env_or("GUIDE_HOST", "127.0.0.1") }
-fn guide_port() -> String    { return guide_env_or("GUIDE_PORT", "8771") }
+fn guide_backend() -> String { return config("GUIDE_BACKEND") }
+fn guide_host() -> String    { return config("GUIDE_HOST") }
+fn guide_port() -> String    { return config("GUIDE_PORT") }
 fn guide_base_url() -> String { return "http://" + guide_host() + ":" + guide_port() }
 
 // guide_healthy — is the guide present and answering? llama-server's /health
@@ -1271,9 +1319,9 @@ fn guide_healthy() -> Bool {
 fn guide_load(tier: String) -> Bool {
     if guide_healthy() { return true }
     let path: String = guide_model_path(tier)
-    let bin: String  = guide_env_or("GUIDE_LLAMA_SERVER_BIN", "llama-server")
-    let ngl: String  = guide_env_or("GUIDE_NGL", "99")
-    let ctx: String  = guide_env_or("GUIDE_CTX", "4096")
+    let bin: String  = config("GUIDE_LLAMA_SERVER_BIN")
+    let ngl: String  = config("GUIDE_NGL")
+    let ctx: String  = config("GUIDE_CTX")
     let logf: String = guide_cache_dir() + "/llama-server." + guide_port() + ".log"
     let cmd: String = bin + " -m '" + path + "' --host " + guide_host() + " --port " + guide_port() + " -c " + ctx + " -ngl " + ngl + " --jinja >> '" + logf + "' 2>&1"
     let pid: String = exec_bg(cmd)
@@ -1669,7 +1717,7 @@ fn route_supersede(method: String, path: String, body: String) -> String {
 // ── Auth ──────────────────────────────────────────────────────────────────────
 
 fn check_auth_ok(method: String, body: String) -> Bool {
-    let key: String = env("ENGRAM_API_KEY")
+    let key: String = config("ENGRAM_API_KEY")
     if str_eq(key, "") { return true }
     // Read-only methods don't require auth. Until http_serve surfaces
     // request headers we can't accept a Bearer token cleanly; mutating
@@ -1932,8 +1980,7 @@ fn handle_request(method: String, path: String, body: String) -> String {
 
 // ── Entry ─────────────────────────────────────────────────────────────────────
 
-let bind_raw: String = env("ENGRAM_BIND")
-let bind_str: String = if str_eq(bind_raw, "") { ":8742" } else { bind_raw }
+let bind_str: String = config("ENGRAM_BIND")
 let port: Int = parse_port(bind_str)
 
 // On startup, try to load any existing snapshot (best effort).