chore(release): v3.11.4 #295

Closed

claude-bot wants to merge 0 commits from release/v3.11.4 into main

claude-bot commented

2026-08-07 06:03:06 +00:00

Collaborator

Cuts v3.11.4 from main at 0a8910e.

Silent failures made visible, and a frontend toolchain that had been building on an end-of-life runtime.

No migrations. BOT_CONTRACT_VERSION stays 1 — nothing touched /api/bot/* this cycle, so the backend and bot images can be upgraded independently. make check-versions passes: app 3.11.4, bot contract v1.

What's in it

Changed

react-router-dom 7 → react-router 8 (#284). The old package is no longer published; this also empties the npm-audit allowlist, so that gate is a real check again rather than a formality.
Frontend now builds and tests on Node 22 (#284). Node 20 went EOL on 2026-04-30, so the production frontend image was being built on an unsupported runtime. CI and both Dockerfile stages move together.

Fixed

Approving a wiki proposal with no working LLM quietly produced a worse entry (#287) — it appended both bodies instead of merging and said nothing. Also covers a misconfigured endpoint, which was previously indistinguishable from an unconfigured one.
A model response cut short at its output cap was silently repaired into a partial result (#293) — twelve proposals could arrive as three, logged as a successful recovery.
Docker builds no longer ship the whole working tree as build context — there was no .dockerignore, so every build sent ~226 MB of node_modules plus full .git history.

Verification so far

Everything in this release is already on main and green:

896 backend tests, 431 frontend tests, ruff check + format (pinned 0.4.4), eslint
frontend-prod, backend-prod and bot-prod images all build
deployed to dev at 26d63a5 and exercised beyond container health: SPA deep-link refresh on every route, manual browser pass, a real LLM call confirming the truncation guard fires correctly and doesn't misfire, and a live check that generated summaries come back as prose

The only commit not yet on dev is 0a8910e (#287), which adds two log lines and no user-facing behaviour.

Self-hosters building their own images pick up Node 22 automatically — nothing to change by hand.

Cuts **v3.11.4** from `main` at `0a8910e`. _Silent failures made visible, and a frontend toolchain that had been building on an end-of-life runtime._ **No migrations. `BOT_CONTRACT_VERSION` stays 1** — nothing touched `/api/bot/*` this cycle, so the backend and bot images can be upgraded independently. `make check-versions` passes: `app 3.11.4, bot contract v1`. ## What's in it **Changed** - `react-router-dom` 7 → `react-router` 8 (#284). The old package is no longer published; this also empties the npm-audit allowlist, so that gate is a real check again rather than a formality. - Frontend now builds and tests on **Node 22** (#284). Node 20 went EOL on 2026-04-30, so the production frontend image was being built on an unsupported runtime. CI and both Dockerfile stages move together. **Fixed** - Approving a wiki proposal with no working LLM quietly produced a worse entry (#287) — it appended both bodies instead of merging and said nothing. Also covers a *misconfigured* endpoint, which was previously indistinguishable from an unconfigured one. - A model response cut short at its output cap was silently repaired into a partial result (#293) — twelve proposals could arrive as three, logged as a successful recovery. - Docker builds no longer ship the whole working tree as build context — there was no `.dockerignore`, so every build sent ~226 MB of `node_modules` plus full `.git` history. ## Verification so far Everything in this release is already on `main` and green: - 896 backend tests, 431 frontend tests, ruff check + format (pinned 0.4.4), eslint - `frontend-prod`, `backend-prod` and `bot-prod` images all build - deployed to dev at `26d63a5` and exercised beyond container health: SPA deep-link refresh on every route, manual browser pass, a real LLM call confirming the truncation guard fires correctly and doesn't misfire, and a live check that generated summaries come back as prose The only commit not yet on dev is `0a8910e` (#287), which adds two log lines and no user-facing behaviour. Self-hosters building their own images pick up Node 22 automatically — nothing to change by hand.

claude-bot added 1 commit

2026-08-07 06:03:07 +00:00

chore(release): v3.11.4

CI / Backend lint (ruff) (pull_request) Successful in 35s

Details

CI / Bot/backend version sync (pull_request) Successful in 23s

Details

CI / Docker image build (pull_request) Successful in 1m0s

Details

CI / Frontend tests, audit, and build (pull_request) Successful in 3m5s

Details

CI / Bot tests and audit (pull_request) Successful in 3m11s

Details

CI / Backend migration, tests, and audit (pull_request) Successful in 6m58s

Details

9c4c835795

Silent failures made visible, and a frontend toolchain that had been building
on an end-of-life runtime.

No migrations. BOT_CONTRACT_VERSION stays 1 — nothing touched /api/bot/* this
cycle — so the backend and bot images can be upgraded independently.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

rbrooks referenced this pull request from a commit

2026-08-07 06:11:53 +00:00

Merge PR #295: v3.11.4 — silent failures made visible, and Node 22

claude-bot commented

2026-08-07 06:18:44 +00:00

Author

Collaborator

Merged as 6ee2bb9 and tagged v3.11.4; the release published with the changelog body. Closing manually.

Worth recording: the merge API refused this PR with HTTP 422 twice, despite mergeable: true, all six CI / * checks green, required_approvals: 0, and allow_merge_commits: true — i.e. every branch-protection condition satisfied. I merged locally with --no-ff instead, which produced the same merge commit shape as previous releases, and pushed to main. CI is green on the merge.

Since the merge did not go through the API, Forgejo left this PR open rather than marking it merged. No functional difference — the commits are on main — but the "merged" badge will be absent from this one.

If this recurs on the next release, it is worth looking at from the Forgejo side rather than working around it again; a merge endpoint that 422s on a PR meeting every stated condition is a papercut that will keep costing a few minutes each time.

Merged as `6ee2bb9` and tagged `v3.11.4`; the release published with the changelog body. Closing manually. Worth recording: the merge API refused this PR with **HTTP 422** twice, despite `mergeable: true`, all six `CI / *` checks green, `required_approvals: 0`, and `allow_merge_commits: true` — i.e. every branch-protection condition satisfied. I merged locally with `--no-ff` instead, which produced the same merge commit shape as previous releases, and pushed to `main`. CI is green on the merge. Since the merge did not go through the API, Forgejo left this PR open rather than marking it merged. No functional difference — the commits are on `main` — but the "merged" badge will be absent from this one. If this recurs on the next release, it is worth looking at from the Forgejo side rather than working around it again; a merge endpoint that 422s on a PR meeting every stated condition is a papercut that will keep costing a few minutes each time.

claude-bot closed this pull request

2026-08-07 06:18:48 +00:00