fix(auto-recall): thread AbortSignal + BM25-only mode + fix post-timeout rejection

[jlin53882]

Summary

Consolidates two closed/in-conflict PRs into a single reviewable branch:

• PR fix(auto-recall): thread AbortSignal to cancel embedding on timeout #668 (closed — AbortSignal threading): auto-recall timeout now cancels in-flight embedding HTTP calls via AbortController threaded through recallWork → retrieveWithRetry → embedder.embedQuery.
• PR fix: restore CLI compat and unblock auto-recall latency #708 (in conflict — BM25-only mode): auto-recall uses lightweight BM25 retrieval by default, skipping remote embedding latency.
• Blocking fix: Promise.race post-timeout rejection now consumed via .catch() so it does not become an unhandled process rejection.

Commits

1. dc7ec59 fix(auto-recall): thread AbortSignal to cancel embedding on timeout
2. 9f4c86c fix(auto-recall): narrow abort classification to our controller
3. 3dc8644 fix: restore cli compat and unblock auto-recall latency
4. b0d03ec fix(auto-recall): consume post-timeout rejection to avoid unhandled rejection

Maintainer feedback addressed

From	Issue	Status
PR #668 review	Post-timeout rejection needs explicit handling	Fixed in b0d03ec

Rebased onto latest origin/master (47b635d) — no conflicts.

[chatgpt-codex-connector]

💡 Codex Review

memory-lancedb-pro/src/retriever.ts

Line 598 in b0d03ec

try {

Restore catch/finally for retrieve try block

The retrieve() method now opens a try block but never follows it with catch or finally, so TypeScript parsing fails (TS1472) and the plugin cannot load at all. This is a hard runtime blocker for any environment that imports src/retriever.ts.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Pass AbortSignal in the correct retrieval argument slot

This call passes signal as the 6th positional argument, but hybridRetrieval expects source and diagnostics before signal, so the abort signal never reaches embedder.embedQuery(...). In vector/hybrid paths that provide a signal, cancellation will not work and timeouts can still block while remote embedding requests run to completion.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Normalize memory_get paths before workspace boundary check

The boundary check uses join(workspaceDir, relPath) plus startsWith(workspaceDir), which can be bypassed with paths like ../workspaceDir2/... that still share the same prefix string. In compat readFile, this allows reading files outside the configured workspace when relPath is user-controlled.

Useful? React with 👍 / 👎.

[rwmjhb]

Please rebase this branch on current master before continuing review.

The branch is behind current master and currently carries packaging failures plus a mixed diff across import-markdown, auto-recall, and retriever behavior. Rebase first, then split or narrow the PR so the remaining diff is reviewable.

[jlin53882]

@rwmjhb 此PR #746 由 PR #779 取代此PR

[jlin53882]

Thanks for the work on this! I've taken the core signal-threading concept and combined it with a cleaner fallback design (James's approach) in a new PR: #783

What changed vs this PR:

• Embedding API timeout → AbortSignal properly threaded to embedQuery(query, signal) (not just returning undefined)
• BM25 fallback: if embed fails, falls back to BM25 instead of returning nothing (James's design — doesn't change the default behavior)
• Removed the "always use BM25 for auto-recall" approach from PR fix: restore CLI compat and unblock auto-recall latency #708 (it was a quality regression)

This PR #746 had good intent but the implementation had some TS errors. The new PR #783 is a clean rebuild.

[jlin53882]

Thanks for the work on this! I've taken the core signal-threading concept and combined it with a cleaner fallback design (James's approach) in a new PR: #784

What changed vs this PR:

• Embedding API timeout → AbortSignal properly threaded to embedQuery(query, signal) (not just returning undefined)
• BM25 fallback: if embed fails, falls back to BM25 instead of returning nothing (James's design — doesn't change the default behavior)
• Removed the "always use BM25 for auto-recall" approach from PR fix: restore CLI compat and unblock auto-recall latency #708 (it was a quality regression)

This PR #746 had good intent but the implementation had some TS errors. The new PR #784 is a clean rebuild.