Skip to content

chore: bump acp-kernel to 0.0.53 (persist ENOENT whole-cycle retry) - #560

Merged
ranxianglei merged 1 commit into
masterfrom
2026-09-06_kernel-0.0.53
Sep 5, 2026
Merged

ranxianglei merged 1 commit into
masterfrom
2026-09-06_kernel-0.0.53

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What

Bumps the acp-kernel pin from 0.0.52 → 0.0.53 (exact pin per AGENTS.md §2; real npm install so the bundle picks up the new kernel).

What 0.0.53 carries

acp-kernel PR #201 — fix: retry whole write cycle on transient ENOENT/ENOTDIR (CI Windows temp sweep):

  • New isTransientFsError covers EPERM/EBUSY/EACCES/ENOENT/ENOTDIR; the old code rethrew ENOENT immediately.
  • writeInner / flushSync retry the whole cycle per attempt (mkdir moved inside the loop, fresh tmp name, write, rename) — a swept directory is recreated.
  • The spill write gets its own transient retry window.
  • 4 new fault-injection tests; kernel suite 575/575.

This closes the loop on the windows-22 CI flake (data layer): GitHub runners sweep %TEMP% mid-test, killing the tmp file between write and rename. Companion test-layer fix (polling instead of fixed sleeps) already landed in PR #559.

Pre-flight

  • typecheck ✓
  • npm test 1046/1048 (2 known env-dependent fails in plugin-agent.test.ts, unrelated)
  • build ✓
  • verified node_modules/acp-kernel = 0.0.53 before rebuild

Consumes acp-kernel v0.0.53 (PR #201): the whole write cycle (mkdir +
fresh tmp + write + rename) is retried on transient ENOENT/ENOTDIR —
Windows %TEMP% sweeps / AV no longer drop session persists — and the
spill write retries too. Test-side polling shipped in PR #559.
@github-actions

github-actions Bot commented Sep 5, 2026

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-09-06_kernel-0.0.53 (10e6e47)

Option A — Install from npm PR tag (recommended)

npm install -g billion-context@pr-560

Each push to this PR publishes a new version under the pr-560 npm tag.

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr560.tgz
npm install -g package

This comment is automatically updated on each push.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 ework agent · qwen3.8-27b

[bot] 🏷 Reviewing this PR now: verifying the diff against the description, confirming acp-kernel@0.0.53 is live on npm (AGENTS.md §5 cross-repo rule), and re-running pre-flight (typecheck / test / build). Full verdict shortly.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 ework agent · qwen3.8-27b

[bot] 🏷 Review verdict: ✅ LGTM — with one stacking note below. Merging stays human-only per AGENTS.md.

Duplicate screening: only similar item is #375 (bump to 0.0.47, different kernel change) — distinct bump, not a duplicate.

Diff verification (vs master):

Cross-repo rule (§5): npm view acp-kernel version → 0.0.53. Live on npm before consuming ✓.

Pre-flight re-run on branch tip 10e6e47:

  • typecheck ✓
  • build ✓ (dist/index.js 2.76 MB, kernel inlined)
  • bundle spot-check: isTransientFsError present, 5× ENOTDIR references — the release v0.1.46 #201 whole-cycle retry logic is genuinely in the built artifact, not just the lockfile
  • npm install → node_modules/acp-kernel = 0.0.53 confirmed (matches your claim)
  • npm test: 1047/1048. Single fail: resolveClientCommand: codex/claude resolve to themselves (tests/launcher.test.ts) — expects bare codex, gets /usr/bin/codex because this sandbox has codex on PATH. Proven pre-existing: identical failure on master (98/99 there), zero relation to the kernel bump. Note: the 2 env-dependent plugin-agent.test.ts fails you saw didn't reproduce here (0 plugin-agent fails) — env-dependent sets differ per machine; neither category touches this change.

No issues found. Ready for human merge whenever (and decide on #559 ordering).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant