docs(changelog): add the 1.0.0-rc.7 section for the rc.6 fix wave - #1921
Conversation
One consolidated CHANGELOG entry for the fourteen rc.6 sweep findings fixed in this wave (#1885, #1890-#1894, #1897-#1900, #1902-#1905), written as a single PR on purpose: per-PR changelog hunks re-conflict every open branch on every merge (#1545), so no fix PR in the wave touched this file. Records what is NOT fixed as well — #1888 remains the open GA blocker, and the four other sweep findings still open are named in Known issues. Verified with scripts/gen_release_notes.py --tag v1.0.0-rc.7: the section extracts cleanly (6 highlights, 0 breaking, 5 migrations). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6e45d71724
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".
|
|
||
| The rc.6 validation sweep ran the fleet again — fresh installs, in-place | ||
| updates, a CPU-only box and a Vulkan-only box — and filed eighteen findings. | ||
| rc.7 closes fourteen of them. The theme this time is silence: a guard that was |
There was a problem hiding this comment.
The rc.6 sweep registered eighteen findings (#1888–#1905), but this section leaves five of them open: the GA blocker #1888 plus the four findings listed under Known issues (#1889, #1895, #1896, and #1901). Only thirteen sweep findings are closed; #1885 is an additional fix from outside that sweep. Claiming fourteen of the eighteen are closed contradicts the release's own Known issues inventory and overstates the fix wave.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 43e6496. The intro now says rc.7 closes twelve sweep findings outright and retires the lane behind a thirteenth (#1888 via #1923); #1885 and #1883 are recorded as fixes from outside the sweep. Open inventory (#1888 pending validation, #1889, #1893, #1895, #1896, #1901) now sums to eighteen with the closed set.
| - `[dispatcher].stream_total_timeout_s` (default `900`) and | ||
| `[dispatcher].stream_idle_timeout_s` (default `300`) bound how long a relayed | ||
| stream may run in total and how long it may go without producing a chunk. Set | ||
| either to `0` to disable that bound. Both are settable through the settings | ||
| API and documented with the rest of the dispatcher configuration (#1893). |
There was a problem hiding this comment.
At the reviewed commit, DispatcherConfig still has extra="forbid" and defines only prefetch_timeout_s, direct_read_timeout_s, and prefetch_parallel_cap; neither timeout named here exists anywhere outside this new changelog section, nor is either passed into Dispatcher. Consequently, an rc.7 artifact cut from this commit still has unbounded streams, and an operator following these release notes by adding either key to hal0.toml makes configuration validation fail instead of enabling the advertised guard. The implementation commits must precede this release section (the same tree also still contains several other entries described here as fixed).
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 43e6496. PR #1918 (#1893, the stream stall guard that introduces these keys) did not merge — it is still in review with changes requested — so the Added section documenting the keys, the Highlights bullet, the migration note, and the Rollback sentence are all removed, and #1893 is listed under Known issues as still open. The changelog no longer documents configuration that does not exist in the tree.
…ssues, #1923 lane retirement, #1883 bundle proxy The rc.7 section claimed all fourteen fix-wave PRs; #1918 (#1893 stream stall guard) is still in review with changes requested, so its Highlights, Added, Audience, migration and Rollback text moves to Known issues. The merged Vulkan lane retirement (#1923) gets Highlights, a Breaking entry, the #1924 slot-TOML migration, and a rewritten #1888 Known-issues paragraph. The releases.hal0.dev cosign-bundle proxy fix (hal0-web#107) closes #1883 and is recorded under Fixed and Supported upgrades. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
…rd to the release PR #1918 (#1893 stream stall guard) merged to main as 66b2dd4 after this branch's last revision (43e6496) had moved it out of the 1.0.0-rc.7 section into Known issues on the assumption it would miss the tag. Reverse that with current-main-accurate facts: restore the Highlights bullet, the Added section documenting [dispatcher].stream_total_timeout_s (default 900) and stream_idle_timeout_s (default 300) plus the direct_read_timeout_s header-wait clarification from the final review-fix commit, the Audience clause, the Operator migrations "streams now bounded by default" bullet, and the Rollback sentence. Drop #1893 from the Known issues open inventory and recount the section intro: closes thirteen of eighteen sweep findings outright and retires the Vulkan lane behind a fourteenth (was twelve/ thirteenth with #1893 unmerged). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
…ed, #1924 migration automated Four gated fix-wave PRs landed on main after the last changelog pass: #1933 (#1895 ownership rows), #1935 (#1901 updater journal), #1937 (#1889 podman seam image reads), #1934 (#1924 gpu-vulkan slot migration). Move the first three from Known issues into Fixed with user-facing prose, describe the #1924 migration in Breaking and Operator migrations (now automatic on `hal0 update`, no longer a manual slot edit), and recount the intro/Known-issues totals: sixteen of eighteen sweep findings closed, the lane retired behind #1888/#1923, and only #1896 left open (fix in progress, not yet merged).
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
1 participant
What
Adds a single
## [1.0.0-rc.7] — 2026-08-17section toCHANGELOG.mdcovering the whole rc.6 fix wave. No code changes.One entry per user-visible change, each referencing its issue:
bge-reranker-base-q4_k_mdropped from the curated catalogue[dispatcher]keys) end a wedged relayed streamPOST /api/memory/addfails fast when the extraction slot's context is below the floorhal0 bench historyprintsno recordsinstead of nothinghal0composite slot reason + logs, liveuptime_s,agent peerscrashhal0 config editprints its sudo hint instead of a tracebackWhy one PR
Deliberate, and deliberately last: per-PR CHANGELOG hunks re-conflict every open branch on every merge (#1545), which is why no fix PR in this wave touched the file.
Also recorded
Verification
python3 scripts/gen_release_notes.py --tag v1.0.0-rc.7 --channel previewextracts the section cleanly: 6 highlights, 0 breaking, 5 migrations, 15.3 KBRELEASE_NOTES.md.pytest tests/release/test_notes.py tests/docs -q→ 42 passed.###, structured bullets have a complete unindented first physical line, and every cross-reference is an absolute GitHub URL.Out of scope
Choosing and cutting the tag. The heading assumes the next preview tag is
v1.0.0-rc.7; if the cut lands on a different version the heading needs to change with it.🤖 Generated with Claude Code