Skip to content

Commit 5965256

Browse files
authored
docs: record 2.77.0 candidate acceptance and publication (#6561)
* docs: record independent frozen-candidate regression acceptance * docs: preserve failed Windows candidate gate and repair boundary * docs: record 2.77.0 candidate acceptance after unchanged Windows rerun * docs: record 2.77.0 publication outcome
1 parent 87e3156 commit 5965256

4 files changed

Lines changed: 165 additions & 0 deletions

File tree

Lines changed: 63 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,63 @@
1+
# Independent regression acceptance
2+
3+
The integrated source at `0818ea1812a028e1c14cd0b0511b44863407bc52`
4+
(tree `98c6c2f05364c8274d404ae5f6ecc56bb73450d2`) is accepted for the next
5+
complete candidate CI run. Production publication remains pending.
6+
7+
Four fresh-context workers ran in parallel with disjoint temporary state and
8+
shared test locking. All 63 selected original test files executed: **2,423 pass,
9+
0 fail, 0 skip**. Separate current-source GUI verification passed 48 tests;
10+
a separate runner-isolation preflight passed 25. Those counts are not added to
11+
2,423. The candidate GUI build succeeded and its 94 generated artifact hashes
12+
remained unchanged through the manual browser verification.
13+
14+
| Area | Original tests | Additional acceptance |
15+
| --- | ---: | --- |
16+
| Native accounts, catalog and fallback | 990 | Seventeen HTTP scenarios and the actual account CLI, including revoked versus transient credentials, replacement races, hard locks, precise model refusal and committed output. |
17+
| Request bytes, compaction and tool history | 370 | UTF-8 threshold captures, destination and cancellation controls, typed Messages errors, recovery limits, and ordered immutable Ollama tool history. |
18+
| Pairing, identity and existing-format accounting | 363 | Actual CLI minting, HTTP redemption, browser interaction, identity race negatives, capacity refusal, and durable legacy-journal restart/corruption. |
19+
| Provider, platform, CLI and historical fixtures | 700 | Discovery and saved-effort persistence across processes, operator commands, synthetic launcher controls, and eight separately executed original historical files. |
20+
21+
A fresh independent acceptance review found no blocker to final candidate CI.
22+
Two independent visual reviews found only Low wrapping and validation-layout
23+
polish issues, with no observed clipped text or blocked controls. These are
24+
retained follow-ups, not claimed fixes or a complete accessibility audit.
25+
26+
## Evidence limits and failed attempts
27+
28+
Failed harness executions remain failed in the private evidence ledger. They
29+
were not overwritten or included as passing executions. Corrected probes use
30+
source-backed fixture contracts. Two request-probe expectations received a
31+
separate independent adjudication: cancellation prevented another physical send
32+
although the legacy reset-only counter remained zero; a preflighted compaction
33+
stream legitimately returned HTTP 400 rather than the HTTP 200 required for an
34+
already-open Messages stream. Neither result proves complete legacy accounting
35+
or preservation of client-visible partial compaction text.
36+
37+
Manual expiry evidence proves rejection of an expired signed mint capability,
38+
not expiry of an issued pairing code. Current CLI/HTTP/browser pairing passed;
39+
full native SSH enrollment at this candidate was unavailable. Earlier native
40+
join/replacement and teardown evidence remains historical. The named pairing
41+
owners are byte-identical, which does not make that a fresh integrated native run.
42+
43+
Every one of the 20 uniquely named assertions from the earlier broad failed run
44+
maps to a current pass in its original file. The former GUI setup error has
45+
current original-file coverage as well. This does not certify the old broad run
46+
or identify its historical state writer; its original 21-failure/one-error total
47+
is preserved. Complete cross-platform CI is still required.
48+
49+
The original Antigravity Windows/multi-account report remains unproven, the real
50+
launcher scenario remains unavailable, and packaged Windows login behavior is
51+
not established by native helper/parser tests. The canonical spend migration
52+
remains excluded; the landed capacity repair preserves the existing format.
53+
Previously recorded accounting, diagnostic and dependency limitations remain
54+
explicit. No current required product failure was waived by those limitations.
55+
56+
All owned processes, listeners and disposable homes were closed or removed;
57+
source and build identity were rechecked. The artifact validator only proves
58+
schema/source consistency; semantic acceptance comes from the reviewed scenario
59+
matrix and the explicit dispositions above.
60+
61+
Next: run the complete `lane=all` CI at the accepted source SHA, inspect every
62+
requested job, then follow the separate version pre-move, promotion and publication
63+
procedure. This documentation record adds no product changes to that source.
Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
# First complete candidate run: NO-GO
2+
3+
The final candidate check ran once against source
4+
`0818ea1812a028e1c14cd0b0511b44863407bc52` on the immutable
5+
`codex/release-2770-0818-candidate` branch. Run
6+
[37162399902](https://github.com/lidge-jun/opencodex/actions/runs/37162399902),
7+
attempt 1, used `workflow_dispatch` with `lane=all`. Its scope outputs selected
8+
all areas, including native, GUI, desktop packaging, docs and helper matrices.
9+
10+
The result is **NO-GO**: 37 jobs succeeded, Windows shard 8/9 failed, and the
11+
aggregate correctly failed. The separate privacy-only job was intentionally
12+
unrequested; the privacy scan inside `gates` executed successfully.
13+
14+
Linux's four shards, macOS's two shards and full-membership control, and eight
15+
Windows shards passed. macOS widget/app verification and Linux AppImage/deb,
16+
sidecar-keyring and packaged-shell E2E also passed. Those results remain valid
17+
observations of this attempt, not permission to ignore its failed shard.
18+
19+
The reported error is `EPERM` in the temporary-directory teardown of
20+
`server-management-auth.test.ts`, in the self-logout case. The failing batch
21+
reported 140 passes and one failure; later batches in that shard did not run.
22+
The log identifies the recursive removal root, not the particular retained
23+
resource or permission condition. The cause is not established by the error
24+
message or by unchanged owning source relative to the earlier green candidate.
25+
26+
A bounded investigation is collecting causal evidence. No cleanup retry budget,
27+
assertion, test membership, Windows requirement or release gate has been relaxed.
28+
There has been no blind retry, version pre-move, main promotion or publication.
29+
Any separate native diagnostic run is evidence collection only and cannot replace
30+
successful release verification. Production changes require a reviewed,
31+
evidence-backed repair and appropriate current-source regression/CI evidence.
Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
# Candidate acceptance: GO after unchanged Windows rerun
2+
3+
Candidate `0818ea1812a028e1c14cd0b0511b44863407bc52` passed the complete
4+
cross-platform gate. Run
5+
[37162399902](https://github.com/lidge-jun/opencodex/actions/runs/37162399902)
6+
attempt 2 re-ran only the failed `windows 8/9` job and the aggregate, with the
7+
same command, assertions, test membership and removal retry schedule. The shard
8+
passed and the run concluded success: 40 jobs, all nine Windows shards green, and
9+
only the standalone `privacy gate` skipped because manual dispatch does not request
10+
it (the privacy scan inside `gates` passed).
11+
12+
The attempt-1 failure is preserved in `021_candidate_first_run.md`. A separate
13+
non-release Windows diagnostic (run
14+
[37168238608](https://github.com/lidge-jun/opencodex/actions/runs/37168238608),
15+
same runner image `20260925.250.1` and Bun 1.4.0) replayed the failing six-file
16+
batch with instrumentation. The self-logout case's temp home was removed on the
17+
first attempt, so the failure did not reproduce in isolation. That run is
18+
diagnostic evidence only and was never treated as release proof; the diagnostic
19+
branches are not merged.
20+
21+
What remains open: the cause of the intermittent teardown `EPERM` is not
22+
established. The diagnostic replay changed the temp root from the profile's
23+
8.3 short path to the runner temp drive, so it cannot exclude the lexical
24+
alias gap in the Windows ACL reap matcher, where a short-name root does not
25+
match long-name pending reaps. That gap is a real weakness, but nothing shows it
26+
caused this failure. It is a follow-up for the test-cleanup path, and the
27+
release gate did not depend on it.
28+
29+
The candidate is promoted unchanged: the `main` promotion merge `06841165f8`
30+
(#6550) has the same tree as `0818ea1812`, after the version pre-move #6549
31+
opened `dev` at 2.78.0. Publication is recorded in `031_publication_outcome.md`.
Lines changed: 40 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,40 @@
1+
# Publication outcome: 2.77.0 released
2+
3+
OpenCodex 2.77.0 is published as the stable `latest` release from `main`
4+
commit `06841165f884a9176d701310638b2112aca7a514`, whose tree is identical to
5+
the verified candidate `0818ea1812`.
6+
7+
1. Version pre-move #6549 opened `dev` at 2.78.0 (`87e3156339`) after its exact-head
8+
Cross-platform CI and Service lifecycle passed.
9+
2. Promotion #6550 merged as a merge commit through the owner-authorized admin PR
10+
route. As with #6468, only the main-target `enforce-target` rule and CodeQL's
11+
new-alert count against `main` were red; neither gates `release.yml`.
12+
3. Push-event Cross-platform CI
13+
[37169967963](https://github.com/lidge-jun/opencodex/actions/runs/37169967963)
14+
and Service lifecycle 37169967951 succeeded on `06841165f8`.
15+
4. Release workflow
16+
[37170949169](https://github.com/lidge-jun/opencodex/actions/runs/37170949169)
17+
(`version=2.77.0`, `tag=latest`, `dry-run=false`, `expected-sha=06841165f8`)
18+
succeeded in every job: dispatch validation, preflight, five standalone and three
19+
desktop packages, verification, npm publish, release attachment and outcomes.
20+
21+
Published evidence:
22+
23+
- npm `@bitkyc08/[email protected]`, dist-tag `latest`, gitHead `06841165f8`,
24+
shasum `4b0155005e19588bdaab1661eb02479de5546dad`, SLSA v1 provenance (Sigstore
25+
log index 3076198266). The registry served the version about five minutes after
26+
the publish step acknowledged it.
27+
- GitHub release [v2.77.0](https://github.com/lidge-jun/opencodex/releases/tag/v2.77.0),
28+
tag on `06841165f8`, marked Latest, 25 assets (same set shape as v2.76.0):
29+
standalone archives for darwin arm64/x64, linux arm64/x64 and windows x64, the
30+
macOS app/dmg, Linux AppImage/deb and Windows MSI, checksums, updater signatures
31+
and `latest.json`.
32+
- Docs deploy for `06841165f8` succeeded (run 37169967937).
33+
- Install smoke in a throwaway npm prefix with an isolated home: `ocx --version`
34+
reports `opencodex 2.77.0` and `ocx help` runs; the darwin arm64 archive matches
35+
its published SHA-256.
36+
37+
Not verified here: launching the signed desktop installers on real macOS, Windows
38+
and Linux hosts, the in-app updater path from 2.76.0, and the native limits already
39+
recorded for #6220, #6473 and #6502. The intermittent Windows test-teardown
40+
`EPERM` described in `022_candidate_acceptance.md` remains a test-cleanup follow-up.

0 commit comments

Comments
 (0)