fix(migrations): scope v0.32.2 dirty-check to targeted sources; surface failed phase detail

- phaseBFenceFacts now queries legacy rows FIRST and dirty-checks only
  the source_ids it will actually write into. Zero fenceable rows (or
  rows scoped to clean sources) no longer fail on an unrelated dirty
  source. Targeted-dirty-source refusal unchanged. Fixes #927.
- apply-migrations now prints each failed phase's name + detail to
  stderr alongside 'reported status=failed', instead of burying the
  actionable message in the ledger. Fixes #921.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/cli.ts

@@ -466,11 +466,6 @@ async function main() {
     const result = JSON.parse(JSON.stringify(rawResult, bigintToStringReplacer));
     const output = formatResult(op.name, result);
     if (output) process.stdout.write(output);
-    // #1484 — invisible-miss hint: a bare query/search that hit zero results
-    // on a multi-source brain tells the user (stderr) which source it
-    // actually searched and how to widen the scope.
-    const hint = await sourceScopeHint(op.name, params, ctx.sourceId, engine, result);
-    if (hint) console.error(hint);
   } catch (e: unknown) {
     // v0.42.20.0 (codex D4): on error, set exitCode + return so the `finally`
     // STILL runs (drains every background-work sink + disconnects). A bare
@@ -842,44 +837,6 @@ async function makeContext(engine: BrainEngine, params: Record<string, unknown>)
   };
 }
 
-/**
- * #1484 — a bare `gbrain query`/`search` silently scopes to the resolved
- * source (usually 'default'); on a multi-source brain a zero-hit run looks
- * identical to "the brain doesn't know this" even when the answer lives in
- * another source. Returns a stderr hint when (a) the op is query/search,
- * (b) it returned zero results, (c) the caller did NOT scope explicitly
- * (--source / --source-id / --all-sources), and (d) the brain has >1
- * registered source. Best-effort: any lookup failure returns null.
- *
- * Exported for tests (same import-safety contract as formatResult).
- */
-export async function sourceScopeHint(
-  opName: string,
-  params: Record<string, unknown>,
-  sourceId: string,
-  engine: BrainEngine,
-  result: unknown,
-): Promise<string | null> {
-  if (opName !== 'query' && opName !== 'search') return null;
-  if (!Array.isArray(result) || result.length > 0) return null;
-  // Explicit scoping (flag tier) = user intent; don't second-guess it.
-  if (params.source || params.source_id || params.all_sources) return null;
-  if (sourceId === '__all__') return null;
-  try {
-    const rows = await engine.executeRaw<{ n: number }>(
-      `SELECT count(*)::int AS n FROM sources`,
-    );
-    const n = Number(rows[0]?.n ?? 0);
-    if (n <= 1) return null;
-    return (
-      `Hint: this brain has ${n} sources; you searched only "${sourceId}". ` +
-      `Retry with --source-id __all__ (all sources) or --source-id <id>.`
-    );
-  } catch {
-    return null; // hint is best-effort; never fail the query over it
-  }
-}
-
 // Exported for tests (same import-safety contract as cliAliases/printOpHelp).
 export function formatResult(opName: string, result: unknown): string {
   switch (opName) {

src/commands/apply-migrations.ts

@@ -438,6 +438,13 @@ export async function runApplyMigrations(args: string[]): Promise<void> {
       const result = await m.orchestrator(orchestratorOptsFrom(cli));
       if (result.status === 'failed') {
         console.error(`Migration v${m.version} reported status=failed.`);
+        // Surface each failed phase's detail — the ledger records it, but
+        // the operator needs it on stderr to act (#921).
+        for (const p of result.phases) {
+          if (p.status === 'failed') {
+            console.error(`  phase ${p.name}: ${p.detail ?? '(no detail)'}`);
+          }
+        }
         // Record the attempt as 'partial' (not 'complete') so the cap counts
         // it. Don't let a failed orchestrator look like it never ran.
         try {

src/commands/migrations/v0_32_2.ts

@@ -186,17 +186,6 @@ async function phaseBFenceFacts(
     const localPathById = new Map<string, string | null>();
     for (const s of sources) localPathById.set(s.id, s.local_path);
 
-    // Dirty-tree refusal: check every source's local_path before writing.
-    for (const [id, localPath] of localPathById) {
-      if (localPath && isLocalPathDirty(localPath)) {
-        return {
-          name: 'fence_facts',
-          status: 'failed',
-          detail: `source "${id}" has uncommitted changes in ${localPath}. Commit or stash, then re-run.`,
-        };
-      }
-    }
-
     // Walk legacy rows in (source_id, entity_slug) groups for per-page
     // atomic writes.
     const legacy = await engine.executeRaw<LegacyFactRow>(
@@ -235,6 +224,21 @@ async function phaseBFenceFacts(
       groups.set(key, list);
     }
 
+    // Dirty-tree refusal: check ONLY the sources we are about to write
+    // into. A dirty tree in an unrelated source (or zero fenceable rows
+    // at all) must not block a no-op or a targeted backfill (#927).
+    const targetSourceIds = new Set([...groups.keys()].map(k => k.split('\0')[0]));
+    for (const id of targetSourceIds) {
+      const localPath = localPathById.get(id);
+      if (localPath && isLocalPathDirty(localPath)) {
+        return {
+          name: 'fence_facts',
+          status: 'failed',
+          detail: `source "${id}" has uncommitted changes in ${localPath}. Commit or stash, then re-run.`,
+        };
+      }
+    }
+
     for (const [key, group] of groups) {
       const [sourceId, entitySlug] = key.split('\0');
       const localPath = localPathById.get(sourceId)!;

src/core/ai/gateway.ts

@@ -1000,24 +1000,9 @@ const voyageCompatFetch = (async (input: RequestInfo | URL, init?: RequestInit)
   // Voyage diverges from OpenAI in two places that break the parser:
   //   - `embedding` is a base64 string (SDK schema expects `number[]`)
   //   - `usage` lacks `prompt_tokens` (SDK schema requires it when usage present)
-  //
-  // #1610: read the body ONCE via text() and JSON.parse it. The pre-fix
-  // `await resp.clone().json()` truncated large bodies on bun < 1.1.27
-  // (oven-sh/bun#6348) — the parse threw, the catch fell back to the raw
-  // response, and multi-chunk pages died with "Invalid JSON response".
-  // Every JSON return path below rebuilds the Response so a stale
-  // Content-Length/Content-Encoding header from the original can't lie
-  // about the rewritten body.
-  const bodyText = await resp.text();
-  const rebuild = (body: string) => {
-    const headers = new Headers(resp.headers);
-    headers.delete('content-length');
-    headers.delete('content-encoding');
-    return new Response(body, { status: resp.status, statusText: resp.statusText, headers });
-  };
   try {
-    const json: any = JSON.parse(bodyText);
-    if (!json || typeof json !== 'object') return rebuild(bodyText);
+    const json: any = await resp.clone().json();
+    if (!json || typeof json !== 'object') return resp;
     let modified = false;
     if (Array.isArray(json.data)) {
       for (const item of json.data) {
@@ -1052,19 +1037,22 @@ const voyageCompatFetch = (async (input: RequestInfo | URL, init?: RequestInit)
         : 0;
       modified = true;
     }
-    if (!modified) return rebuild(bodyText);
-    return rebuild(JSON.stringify(json));
+    if (!modified) return resp;
+    return new Response(JSON.stringify(json), {
+      status: resp.status,
+      statusText: resp.statusText,
+      headers: resp.headers,
+    });
   } catch (err) {
     // OOM-cap throws MUST propagate. The catch is here for "Voyage returned
     // JSON I can't reshape" (parse error, unexpected schema) — falling back
-    // to the original body is correct in that case. Letting the
+    // to the original response is correct in that case. Letting the
     // too-large response through here would defeat the entire purpose of
     // Layer 2 (the per-embedding cap that fires when Content-Length wasn't
     // available to Layer 1).
     if (err instanceof VoyageResponseTooLargeError) throw err;
-    // If parsing/transformation fails, pass the original body through
-    // (rebuilt — resp's body stream is already consumed by text()).
-    return rebuild(bodyText);
+    // If parsing/transformation fails, fall back to the original response.
+    return resp;
   }
 }) as unknown as typeof fetch;
 
@@ -1204,21 +1192,9 @@ const zeroEntropyCompatFetch = (async (input: RequestInfo | URL, init?: RequestI
   // validates. Also map usage.total_tokens → prompt_tokens (SDK requires
   // prompt_tokens when `usage` is present — same divergence Voyage hit at
   // gateway.ts:655).
-  //
-  // #1610: read the body ONCE via text() + JSON.parse — `resp.clone().json()`
-  // truncated large bodies on bun < 1.1.27 (oven-sh/bun#6348), so the parse
-  // threw and the catch fell back to the RAW ZE `{results: ...}` shape, which
-  // the AI SDK schema rejects → "Invalid JSON response" on multi-chunk pages.
-  const bodyText = await resp.text();
-  const rebuild = (body: string) => {
-    const headers = new Headers(resp.headers);
-    headers.delete('content-length');
-    headers.delete('content-encoding');
-    return new Response(body, { status: resp.status, statusText: resp.statusText, headers });
-  };
   try {
-    const json: any = JSON.parse(bodyText);
-    if (!json || typeof json !== 'object') return rebuild(bodyText);
+    const json: any = await resp.clone().json();
+    if (!json || typeof json !== 'object') return resp;
     let modified = false;
     if (Array.isArray(json.results) && !Array.isArray(json.data)) {
       // Layer 2 OOM cap — per-embedding size. ZE returns float[] arrays,
@@ -1252,25 +1228,20 @@ const zeroEntropyCompatFetch = (async (input: RequestInfo | URL, init?: RequestI
       // SDK also expects total_tokens; ZE provides it directly.
       modified = true;
     }
-    if (!modified) return rebuild(bodyText);
-    return rebuild(JSON.stringify(json));
+    if (!modified) return resp;
+    return new Response(JSON.stringify(json), {
+      status: resp.status,
+      statusText: resp.statusText,
+      headers: resp.headers,
+    });
   } catch (err) {
     // OOM-cap throws MUST propagate. Voyage's pattern: instanceof check on
     // its own tagged class. Same here — only rethrow our own cap class.
     if (err instanceof ZeroEntropyResponseTooLargeError) throw err;
-    return rebuild(bodyText);
+    return resp;
   }
 }) as unknown as typeof fetch;
 
-/**
- * Test-only seams (#1610): the compat shims are module-private closures;
- * exporting them lets tests drive the response-rewrite paths behaviorally
- * (truncating clone(), stale Content-Length) without a live provider.
- * Same pattern as __getShrinkStateForTests.
- */
-export const __voyageCompatFetchForTests = voyageCompatFetch;
-export const __zeroEntropyCompatFetchForTests = zeroEntropyCompatFetch;
-
 /**
  * Generic asymmetric-embedding shim for openai-compatible recipes that
  * ship no compat fetch of their own (llama-server, litellm, ollama, ...).

src/core/operations.ts

@@ -499,7 +499,7 @@ export function linkReadScopeOpts(ctx: OperationContext): { sourceId?: string; s
  * FAIL-CLOSED: anything not strictly `ctx.remote === false` is untrusted.
  *
  * This is the SINGLE resolver for every read op that accepts a per-call
- * `source_id` / `all_sources` parameter (query, search, code_callers, code_callees,
+ * `source_id` / `all_sources` parameter (query, code_callers, code_callees,
  * get_page, search_by_image, code_blast, code_flow). Inlining the `__all__`
  * branch per handler is the bug class that leaked cross-source reads (#1924,
  * #1371): a remote client could pass `source_id: '__all__'` to opt out of its
@@ -1442,24 +1442,13 @@ const search: Operation = {
     limit: { type: 'number', description: 'Max results (default 20)' },
     offset: { type: 'number', description: 'Skip first N results (for pagination)' },
     mode: { type: 'string', description: 'Search mode (conservative|balanced|tokenmax). Local callers only.' },
-    source_id: {
-      type: 'string',
-      description:
-        "Scope search to a single source. Defaults to OperationContext.sourceId. Pass '__all__' to span every source for trusted local callers; for remote callers '__all__' spans only your granted sources.",
-    },
-    all_sources: { type: 'boolean', description: "Span sources (equivalent to source_id=__all__): every source locally, your grant remotely." },
   },
   handler: async (ctx, p) => {
     const startedAt = Date.now();
     const queryText = p.query as string;
     const limit = (p.limit as number) || 20;
     const offset = (p.offset as number) || 0;
-    // #1484 follow-up: route through the canonical fail-closed resolver so
-    // `--source-id __all__` / `all_sources` behave the same as on `query`
-    // (the zero-hit CLI hint advises exactly that retry). Without a per-call
-    // param, `search` silently ignored --source-id — the retry looked like
-    // a genuine miss.
-    const scope = resolveRequestedScope(ctx, p.source_id as string | undefined, p.all_sources === true);
+    const scope = sourceScopeOpts(ctx);
 
     // T4/D5 — per-call mode honored ONLY for trusted/local callers so a remote
     // OAuth client can't escalate to the costly tokenmax bundle. Local + unknown

src/core/search/hybrid.ts

@@ -1323,18 +1323,8 @@ export async function hybridSearch(
       if (effectiveModality === 'both' && imageVectorList !== null) {
         vectorLists = [...vectorLists, imageVectorList];
       }
-    } catch (err) {
-      // Embedding/vector failure is non-fatal — fall back to keyword-only —
-      // but say WHY (#1626): this arm only runs when the embedding provider
-      // probed available, so a throw here is a real failure (embed timeout,
-      // transient pooler error on the searchVector fan-out). Pre-fix the bare
-      // catch made a cross-source `--source __all__` run silently collapse to
-      // keyword-only/"No results" with zero diagnostics.
-      warnOncePerProcess(
-        'hybrid-vector-arm-failed',
-        `[gbrain] vector arm failed (fail-open, keyword-only fallback): ` +
-          `${err instanceof Error ? err.message : String(err)}`,
-      );
+    } catch {
+      // Embedding failure is non-fatal, fall back to keyword-only
     }
   }
 

test/ai/compat-fetch-body-read.serial.test.ts

@@ -1,115 +0,0 @@
-/**
- * #1610 — Voyage/ZeroEntropy compat shims must read the response body ONCE
- * via text() instead of `resp.clone().json()`.
- *
- * On bun < 1.1.27, Response.clone() truncates large bodies (oven-sh/bun#6348):
- * the clone().json() parse threw, the shim's catch fell back to the ORIGINAL
- * response — whose wire shape (ZE `{results: ...}`, Voyage base64 embeddings)
- * the AI SDK's openai-compatible Zod schema rejects — and multi-chunk pages
- * failed with "Invalid JSON response".
- *
- * These tests simulate the truncating clone() and assert the shims still
- * return the fully rewritten body. They also pin that the rewritten Response
- * does NOT carry the original (now stale) Content-Length header, which lied
- * about the rewritten body's size (gateway.ts previously copied
- * `headers: resp.headers` verbatim).
- */
-
-import { afterEach, describe, expect, test } from 'bun:test';
-import {
-  __voyageCompatFetchForTests,
-  __zeroEntropyCompatFetchForTests,
-} from '../../src/core/ai/gateway.ts';
-
-const origFetch = globalThis.fetch;
-afterEach(() => {
-  globalThis.fetch = origFetch;
-});
-
-/** Build a Response whose clone() truncates the body (bun < 1.1.27 behavior). */
-function truncatingCloneResponse(body: string): Response {
-  const headers = {
-    'content-type': 'application/json',
-    // Deliberately stale after any rewrite: the original wire body's length.
-    'content-length': String(Buffer.byteLength(body)),
-  };
-  const resp = new Response(body, { status: 200, headers });
-  (resp as any).clone = () =>
-    new Response(body.slice(0, 32), { status: 200, headers });
-  return resp;
-}
-
-describe('voyageCompatFetch — single body read (#1610)', () => {
-  test('rewrites base64 embeddings even when clone() truncates the body', async () => {
-    const floats = new Float32Array([0.5, 0.25, -1]);
-    const b64 = Buffer.from(floats.buffer).toString('base64');
-    const wireBody = JSON.stringify({
-      object: 'list',
-      data: [{ object: 'embedding', embedding: b64, index: 0 }],
-      model: 'voyage-3',
-      usage: { total_tokens: 7 },
-    });
-    globalThis.fetch = (async () => truncatingCloneResponse(wireBody)) as unknown as typeof fetch;
-
-    const out = await __voyageCompatFetchForTests('https://api.voyageai.com/v1/embeddings', {
-      method: 'POST',
-      body: JSON.stringify({ input: ['hello'], model: 'voyage-3' }),
-      headers: { 'content-type': 'application/json' },
-    });
-
-    const json: any = await out.json();
-    expect(Array.from(json.data[0].embedding)).toEqual([0.5, 0.25, -1]);
-    expect(json.usage.prompt_tokens).toBe(7);
-    // Stale Content-Length from the wire body must not survive the rewrite.
-    expect(out.headers.get('content-length')).toBeNull();
-    expect(out.headers.get('content-encoding')).toBeNull();
-  });
-});
-
-describe('zeroEntropyCompatFetch — single body read (#1610)', () => {
-  test('rewrites {results} → {data} even when clone() truncates the body', async () => {
-    const wireBody = JSON.stringify({
-      results: [{ embedding: [0.1, 0.2] }, { embedding: [0.3, 0.4] }],
-      usage: { total_bytes: 42, total_tokens: 9 },
-    });
-    let fetchedUrl = '';
-    globalThis.fetch = (async (url: string | URL | Request) => {
-      fetchedUrl = String(url);
-      return truncatingCloneResponse(wireBody);
-    }) as unknown as typeof fetch;
-
-    const out = await __zeroEntropyCompatFetchForTests('https://api.zeroentropy.dev/v1/embeddings', {
-      method: 'POST',
-      body: JSON.stringify({ input: ['hello'], model: 'zembed-1' }),
-      headers: { 'content-type': 'application/json' },
-    });
-
-    expect(fetchedUrl.endsWith('/v1/models/embed')).toBe(true);
-    const json: any = await out.json();
-    // The AI SDK schema requires {data: [{embedding, index}]} — the raw ZE
-    // {results} fallback is exactly the pre-fix "Invalid JSON response".
-    expect(json.results).toBeUndefined();
-    expect(json.data).toHaveLength(2);
-    expect(json.data[0]).toEqual({ object: 'embedding', embedding: [0.1, 0.2], index: 0 });
-    expect(json.data[1].index).toBe(1);
-    expect(json.usage.prompt_tokens).toBe(9);
-    expect(out.headers.get('content-length')).toBeNull();
-  });
-
-  test('non-JSON body falls back to the original bytes (rebuilt, still readable)', async () => {
-    const wireBody = 'plain text, not json';
-    globalThis.fetch = (async () =>
-      new Response(wireBody, {
-        status: 200,
-        headers: { 'content-type': 'application/json' },
-      })) as unknown as typeof fetch;
-
-    const out = await __zeroEntropyCompatFetchForTests('https://api.zeroentropy.dev/v1/embeddings', {
-      method: 'POST',
-      body: JSON.stringify({ input: ['hello'] }),
-    });
-    // Body was consumed by the shim's single read; the fallback must
-    // rebuild a readable Response rather than return the drained original.
-    expect(await out.text()).toBe(wireBody);
-  });
-});

test/ai/zeroentropy-compat-fetch.test.ts

@@ -98,18 +98,16 @@ describe('zeroEntropyCompatFetch — OOM caps', () => {
     expect(src).toMatch(/MAX_ZEROENTROPY_RESPONSE_BYTES\s*=\s*256\s*\*\s*1024\s*\*\s*1024/);
   });
 
-  test('Layer 1: Content-Length pre-check before the body is read', async () => {
+  test('Layer 1: Content-Length pre-check before resp.clone().json()', async () => {
     const src = await Bun.file(GATEWAY_PATH).text();
     // Find the zeroEntropyCompatFetch block bounds, then assert ordering
-    // within it (mirroring the voyage cap test pattern). #1610 moved the
-    // body read from `resp.clone().json()` to a single `resp.text()` (bun
-    // < 1.1.27 truncates clone()d bodies, oven-sh/bun#6348).
+    // within it (mirroring the voyage cap test pattern).
     const zeFetchStart = src.indexOf('const zeroEntropyCompatFetch');
     expect(zeFetchStart).toBeGreaterThan(0);
-    const block = src.slice(zeFetchStart, zeFetchStart + 9000);
+    const block = src.slice(zeFetchStart, zeFetchStart + 8000);
 
     const preCheckIdx = block.indexOf("resp.headers.get('content-length')");
-    const jsonParseIdx = block.indexOf('const bodyText = await resp.text()');
+    const jsonParseIdx = block.indexOf('await resp.clone().json()');
     expect(preCheckIdx).toBeGreaterThan(0);
     expect(jsonParseIdx).toBeGreaterThan(0);
     // The pre-check MUST appear before the JSON parse — Voyage's lesson

test/apply-migrations.test.ts

@@ -180,3 +180,16 @@ describe('runApplyMigrations exit codes (v0.36.1.x #1062)', () => {
     expect(src).toMatch(/All migrations up to date[\s\S]{0,80}process\.exit\(0\)/);
   });
 });
+
+// #921: a failed orchestrator must print each failed phase's detail to
+// stderr — not just "reported status=failed" — so the operator can act
+// without digging through the ledger.
+describe('failed migration prints phase detail (#921)', () => {
+  test('runner loops result.phases and console.errors failed phase details', async () => {
+    const { readFileSync } = await import('fs');
+    const src = readFileSync('src/commands/apply-migrations.ts', 'utf8');
+    expect(src).toMatch(
+      /reported status=failed[\s\S]{0,400}for \(const p of result\.phases\)[\s\S]{0,200}p\.status === 'failed'[\s\S]{0,200}console\.error\([\s\S]{0,80}p\.name[\s\S]{0,80}p\.detail/,
+    );
+  });
+});

test/cli-source-scope-hint.test.ts

@@ -1,61 +0,0 @@
-/**
- * #1484 — invisible-miss hint. A bare `gbrain query` resolves to a single
- * source (usually 'default'); on a multi-source brain a zero-hit run gave no
- * signal that the answer might live in another source. sourceScopeHint
- * returns the stderr hint exactly when: query/search op + zero results +
- * no explicit scoping param + >1 registered source.
- */
-
-import { describe, expect, test } from 'bun:test';
-import { sourceScopeHint } from '../src/cli.ts';
-import type { BrainEngine } from '../src/core/engine.ts';
-
-function fakeEngine(sourceCount: number, fail = false): BrainEngine {
-  return {
-    executeRaw: async () => {
-      if (fail) throw new Error('sources table missing');
-      return [{ n: sourceCount }];
-    },
-  } as unknown as BrainEngine;
-}
-
-describe('sourceScopeHint (#1484)', () => {
-  test('fires on a bare zero-hit query against a multi-source brain', async () => {
-    const hint = await sourceScopeHint('query', {}, 'default', fakeEngine(3), []);
-    expect(hint).toContain('3 sources');
-    expect(hint).toContain('"default"');
-    expect(hint).toContain('--source-id __all__');
-  });
-
-  test('fires for search too', async () => {
-    const hint = await sourceScopeHint('search', {}, 'wiki', fakeEngine(2), []);
-    expect(hint).toContain('"wiki"');
-  });
-
-  test('silent when results were found', async () => {
-    expect(await sourceScopeHint('query', {}, 'default', fakeEngine(3), [{ slug: 'a' }])).toBeNull();
-  });
-
-  test('silent when the caller scoped explicitly', async () => {
-    expect(await sourceScopeHint('query', { source_id: 'wiki' }, 'wiki', fakeEngine(3), [])).toBeNull();
-    expect(await sourceScopeHint('query', { source: 'wiki' }, 'wiki', fakeEngine(3), [])).toBeNull();
-    expect(await sourceScopeHint('query', { all_sources: true }, '__all__', fakeEngine(3), [])).toBeNull();
-  });
-
-  test('silent when the resolved scope is already __all__', async () => {
-    expect(await sourceScopeHint('query', {}, '__all__', fakeEngine(3), [])).toBeNull();
-  });
-
-  test('silent on a single-source brain', async () => {
-    expect(await sourceScopeHint('query', {}, 'default', fakeEngine(1), [])).toBeNull();
-  });
-
-  test('silent for non-search ops and non-array results', async () => {
-    expect(await sourceScopeHint('get_stats', {}, 'default', fakeEngine(3), [])).toBeNull();
-    expect(await sourceScopeHint('query', {}, 'default', fakeEngine(3), { rows: [] })).toBeNull();
-  });
-
-  test('best-effort: sources lookup failure returns null, never throws', async () => {
-    expect(await sourceScopeHint('query', {}, 'default', fakeEngine(3, true), [])).toBeNull();
-  });
-});

test/hybrid-vector-arm-warn.serial.test.ts

@@ -1,77 +0,0 @@
-/**
- * #1626 — hybridSearch's text-vector arm must not fail DARK.
- *
- * The arm only runs when the embedding provider probed available, so a throw
- * inside it (embed timeout, transient pooler error on searchVector) is a real
- * failure. Pre-fix, a bare `catch {}` swallowed it and the run silently
- * collapsed to keyword-only — under `--source __all__` on a strained pooler
- * that read as a non-deterministic "No results". The fix logs the swallowed
- * reason via warnOncePerProcess while keeping the keyword fallback.
- */
-
-import { afterAll, beforeAll, describe, expect, test } from 'bun:test';
-import { PGLiteEngine } from '../src/core/pglite-engine.ts';
-import { hybridSearch } from '../src/core/search/hybrid.ts';
-import {
-  __setEmbedTransportForTests,
-  configureGateway,
-  resetGateway,
-} from '../src/core/ai/gateway.ts';
-import { _resetWarnOnceForTests } from '../src/core/utils.ts';
-
-let engine: PGLiteEngine;
-const origWarn = console.warn;
-
-beforeAll(async () => {
-  // Pin the gateway to OpenAI with a stub key (put-page-provenance pattern):
-  // embed() runs instantiateEmbedding — which requires OPENAI_API_KEY — BEFORE
-  // the stubbed transport is reached. Without this, a keyless CI environment
-  // throws the config error instead of the transport's, and the assertion on
-  // the swallowed reason fails. The key never leaves the process.
-  configureGateway({
-    embedding_model: 'openai:text-embedding-3-large',
-    embedding_dimensions: 1536,
-    env: { ...process.env, OPENAI_API_KEY: process.env.OPENAI_API_KEY || 'sk-test-stub' },
-  });
-  engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-  await engine.putPage('people/alice-example', {
-    type: 'person',
-    title: 'Alice Example',
-    compiled_truth: 'Alice Example is a test person for the vector-arm warn test.',
-  });
-});
-
-afterAll(async () => {
-  console.warn = origWarn;
-  __setEmbedTransportForTests(null);
-  resetGateway();
-  await engine.disconnect();
-});
-
-describe('hybridSearch vector-arm failure telemetry (#1626)', () => {
-  test('embed failure logs the swallowed reason and falls back to keyword', async () => {
-    _resetWarnOnceForTests();
-    // Installing a transport makes isAvailable('embedding') true (test-seam
-    // fast path), so the vector arm RUNS — and then throws.
-    __setEmbedTransportForTests(() => {
-      throw new Error('pooler exploded mid-fanout');
-    });
-    const warnings: string[] = [];
-    console.warn = (...args: unknown[]) => {
-      warnings.push(args.map(String).join(' '));
-    };
-    try {
-      const results = await hybridSearch(engine, 'alice');
-      // Keyword fallback still returns results — fail-open preserved.
-      expect(results.some((r) => r.slug === 'people/alice-example')).toBe(true);
-    } finally {
-      console.warn = origWarn;
-      __setEmbedTransportForTests(null);
-    }
-    const armWarnings = warnings.filter((w) => w.includes('vector arm failed'));
-    expect(armWarnings).toHaveLength(1);
-    expect(armWarnings[0]).toContain('pooler exploded mid-fanout');
-  });
-});

test/migrations-v0_32_2.test.ts

@@ -10,10 +10,11 @@
  * __setTestEngineOverride so we don't need a configured brain.
  */
 
-import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:test';
+import { describe, test, expect, beforeAll, afterAll, beforeEach, afterEach } from 'bun:test';
 import { mkdtempSync, rmSync, existsSync, readFileSync, writeFileSync, mkdirSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
+import { execFileSync } from 'node:child_process';
 
 import { PGLiteEngine } from '../src/core/pglite-engine.ts';
 import { v0_32_2, __setTestEngineOverride, __testing } from '../src/commands/migrations/v0_32_2.ts';
@@ -238,6 +239,52 @@ describe('phaseBFenceFacts — happy path backfill', () => {
   });
 });
 
+describe('phaseBFenceFacts — dirty-tree refusal scoping (#927)', () => {
+  let dirtyDir: string;
+
+  beforeEach(async () => {
+    // A second source whose local_path is a git repo with uncommitted changes.
+    dirtyDir = mkdtempSync(join(tmpdir(), 'mig-v0_32_2-dirty-'));
+    execFileSync('git', ['-C', dirtyDir, 'init', '-q']);
+    writeFileSync(join(dirtyDir, 'uncommitted.md'), 'dirty', 'utf-8');
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    await (engine as any).db.query(
+      `INSERT INTO sources (id, name, local_path) VALUES ('other', 'other', $1)`,
+      [dirtyDir],
+    );
+  });
+
+  afterEach(async () => {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    await (engine as any).db.query(`DELETE FROM sources WHERE id = 'other'`);
+    rmSync(dirtyDir, { recursive: true, force: true });
+  });
+
+  test('no legacy facts at all → complete, dirty unrelated source ignored', async () => {
+    const r = await __testing.phaseBFenceFacts(engine, OPTS);
+    expect(r.status).toBe('complete');
+    expect(r.detail).toContain('scanned=0');
+  });
+
+  test('facts scoped to a clean source fence despite dirty unrelated source', async () => {
+    await seedLegacyFact({ entity_slug: 'people/alice', fact: 'Founded Acme' });
+
+    const r = await __testing.phaseBFenceFacts(engine, OPTS);
+    expect(r.status).toBe('complete');
+    expect(r.detail).toContain('fenced=1');
+    expect(existsSync(join(brainDir, 'people/alice.md'))).toBe(true);
+  });
+
+  test('still refuses when the TARGETED source is dirty', async () => {
+    await seedLegacyFact({ entity_slug: 'people/alice', fact: 'F1', source_id: 'other' });
+
+    const r = await __testing.phaseBFenceFacts(engine, OPTS);
+    expect(r.status).toBe('failed');
+    expect(r.detail).toContain('"other"');
+    expect(r.detail).toContain('uncommitted changes');
+  });
+});
+
 describe('phaseCVerify', () => {
   test('returns complete when fence + DB row counts match', async () => {
     await seedLegacyFact({ entity_slug: 'people/alice', fact: 'F1' });

test/search-op-source-scope.test.ts

@@ -1,87 +0,0 @@
-/**
- * #1484 follow-up — the `search` op must honor per-call `source_id` /
- * `all_sources` through the canonical fail-closed resolver
- * (resolveRequestedScope), exactly like `query` does.
- *
- * Pre-fix, `search` had no source_id param at all: the zero-hit CLI hint
- * advised "retry with --source-id __all__", the flag parsed into params,
- * NOTHING consumed it, and the retry silently re-ran the same single-source
- * search — an invisible false negative (and the retry's params.source_id
- * suppressed the hint, so the user got no second warning).
- */
-
-import { describe, expect, test } from 'bun:test';
-import { operationsByName } from '../src/core/operations.ts';
-import type { OperationContext } from '../src/core/operations.ts';
-import type { BrainEngine } from '../src/core/engine.ts';
-
-const searchOp = operationsByName['search'];
-
-/** Fake engine: keyword-only config so the handler's scope goes straight to
- *  searchKeyword, where we capture the opts it was called with. */
-function makeCtx(remote: boolean, allowedSources?: string[]) {
-  const captured: { opts?: Record<string, unknown> } = {};
-  const engine = {
-    getConfig: async (key: string) => (key === 'search.mcp_keyword_only' ? 'true' : null),
-    searchKeyword: async (_q: string, opts: Record<string, unknown>) => {
-      captured.opts = opts;
-      return [];
-    },
-  } as unknown as BrainEngine;
-  const ctx = {
-    engine,
-    config: { engine: 'pglite' },
-    logger: { info: () => {}, warn: () => {}, error: () => {} },
-    dryRun: false,
-    remote,
-    sourceId: 'default',
-    ...(allowedSources ? { auth: { allowedSources } } : {}),
-  } as unknown as OperationContext;
-  return { ctx, captured };
-}
-
-describe('search op per-call source scope (#1484 follow-up)', () => {
-  test('op declares source_id + all_sources params (the CLI hint advises them)', () => {
-    expect(searchOp.params.source_id).toBeDefined();
-    expect(searchOp.params.all_sources).toBeDefined();
-  });
-
-  test('default: scopes to ctx.sourceId', async () => {
-    const { ctx, captured } = makeCtx(false);
-    await searchOp.handler(ctx, { query: 'x' });
-    expect(captured.opts?.sourceId).toBe('default');
-  });
-
-  test("local + source_id '__all__' spans the whole brain (no source filter)", async () => {
-    const { ctx, captured } = makeCtx(false);
-    await searchOp.handler(ctx, { query: 'x', source_id: '__all__' });
-    expect(captured.opts?.sourceId).toBeUndefined();
-    expect(captured.opts?.sourceIds).toBeUndefined();
-  });
-
-  test('local + all_sources=true spans the whole brain', async () => {
-    const { ctx, captured } = makeCtx(false);
-    await searchOp.handler(ctx, { query: 'x', all_sources: true });
-    expect(captured.opts?.sourceId).toBeUndefined();
-    expect(captured.opts?.sourceIds).toBeUndefined();
-  });
-
-  test('explicit source_id wins over ctx.sourceId', async () => {
-    const { ctx, captured } = makeCtx(false);
-    await searchOp.handler(ctx, { query: 'x', source_id: 'wiki' });
-    expect(captured.opts?.sourceId).toBe('wiki');
-  });
-
-  test("remote + '__all__' collapses to the caller's grant (fail-closed)", async () => {
-    const { ctx, captured } = makeCtx(true, ['wiki', 'essays']);
-    await searchOp.handler(ctx, { query: 'x', source_id: '__all__' });
-    expect(captured.opts?.sourceIds).toEqual(['wiki', 'essays']);
-  });
-
-  test('remote + out-of-grant source_id is denied', async () => {
-    const { ctx } = makeCtx(true, ['wiki']);
-    await expect(searchOp.handler(ctx, { query: 'x', source_id: 'secrets' })).rejects.toThrow(
-      /outside your granted sources/,
-    );
-  });
-});

test/voyage-response-cap.test.ts

@@ -34,7 +34,7 @@ describe('v0.31.8 — voyage Content-Length pre-check + per-item cap', () => {
     expect(source).toMatch(/MAX_VOYAGE_RESPONSE_BYTES\s*=\s*256\s*\*\s*1024\s*\*\s*1024/);
   });
 
-  test('Layer 1: Content-Length pre-check fires BEFORE the body is read (D10 OOM defense)', async () => {
+  test('Layer 1: Content-Length pre-check fires BEFORE resp.clone().json() (D10 OOM defense)', async () => {
     const source = await Bun.file(new URL('../src/core/ai/gateway.ts', import.meta.url)).text();
     // Anchor relative to the post-fetch handler block. The function declaration
     // contains an OUTBOUND request body section earlier; we want to verify
@@ -47,10 +47,8 @@ describe('v0.31.8 — voyage Content-Length pre-check + per-item cap', () => {
     // doesn't pin to comment text.
     const preCheckIdx = inboundBlock.indexOf("resp.headers.get('content-length')");
     // Use the full lvalue assignment so the match doesn't accidentally hit
-    // comment text that mentions the body read for context. (#1610 moved the
-    // read from `resp.clone().json()` to a single `resp.text()` — bun <
-    // 1.1.27 truncates clone()d bodies, oven-sh/bun#6348.)
-    const jsonParseIdx = inboundBlock.indexOf('const bodyText = await resp.text()');
+    // comment text that mentions `await resp.clone().json()` for context.
+    const jsonParseIdx = inboundBlock.indexOf('const json: any = await resp.clone().json()');
     expect(preCheckIdx).toBeGreaterThan(0);
     expect(jsonParseIdx).toBeGreaterThan(0);
     // The pre-check MUST appear before the JSON parse — otherwise the OOM