Skip to content

Commit 2eeb091

Browse files
sepo-engFelicityclaude
authored
ci(fork): make CI green-means-green -- blacksmith sweep, unit reds fixed (#11)
* ci(fork): retarget/disable blacksmith runners, stabilize windows/linux unit reds TKT-336, part 1: work-in-progress commit to get real CI evidence before finalizing FORK.md. Squashed/reworded before merge. Blacksmith-runner sweep (18+ workflow files queued forever, no Blacksmith GitHub App on this org): - test.yml e2e job: retargeted to ubuntu-latest/windows-latest. Real Playwright coverage, and the literal cause of "held open" workflow runs blocking rerun-failed-jobs for every job in the same run. - storybook.yml, notify-discord.yml: retargeted (real, cheap checks). - nix-eval.yml, nix-hashes.yml, generate.yml, docs-update.yml, docs-locale-sync.yml, publish.yml, publish-github-action.yml, publish-vscode.yml, containers.yml, release-github-action.yml, beta.yml: auto-triggers disabled (workflow_dispatch kept as escape hatch). Not this fork's concern (nix packaging, docs sync) or a safety choice (real npm/container/marketplace publish pipelines with write credentials -- an internal redesign fork should not auto-publish under the org's name). - pr-management.yml (check-duplicates), review.yml, opencode.yml: job-level `if: false`, not just runner-blocked -- also missing OPENCODE_API_KEY (already documented in FORK.md for review.yml). - duplicate-issues.yml, triage.yml: untouched, already fully inert (has_issues: false on this repo, issues: events can never fire). - stats.yml: untouched, already self-excludes via `if: github.repository == 'anomalyco/opencode'`. unit (linux): the anomalyco#27371 fast-fail regression test skipped (not widened -- widening the 15s assertion would mask the specific failure mode feedback anomalyco#136 suspects, not just add margin), comment links anomalyco#136 and the un-skip condition. unit (windows) mode A: packages/core's test script had no --timeout override at all (bun's raw 5000ms default), unlike opencode/tui which already use --timeout 30000. Every test FORK.md documented as a mode-A failure (Ripgrep, i18n parity, Git worktrees/trees, LocationServiceMap, MoveSession) lives in packages/core -- confirmed via grep, not assumed. Added --timeout 30000, matching the established fork convention elsewhere. Not yet verified against real CI (that's this commit's purpose) or reflected in FORK.md's known-red table -- both follow once 2-3 windows runs confirm whether the core timeout fix alone stabilizes it. Co-Authored-By: Claude <[email protected]> * ci(fork): skip windows unit tests -- timeout raise exposed a real Ripgrep bug First real CI run of TKT-336's changes showed the packages/core timeout fix (30s override) genuinely stabilized the mode-A "varying subset" flakiness -- i18n parity, Git worktrees/trees, LocationServiceMap, and MoveSession all passed. But Ripgrep's 4 tests still failed, now timing out at exactly 30000ms instead of ~5000ms, with "Ripgrep.Error: ripgrep execution failed" in the log -- a real, deterministic execution failure on windows-latest, not a timing margin issue that more rope would fix. RipgrepBinary's Windows path downloads a zip and extracts it via PowerShell at first use (packages/core/src/ripgrep/binary.ts) -- root-causing that is real production-code debugging, out of scope for a CI-hygiene ticket. Filed as feedback anomalyco#152 with the full evidence. Per the ticket's own stated fallback ("skip the windows unit job entirely with a stated rationale if timeouts alone do not stabilize it"), the "Run unit tests" step is now Linux-only. check:generated and test:httpapi were already Linux-only; nothing else runs on Windows in this job now, so it reports a clean pass instead of vanishing or red. Co-Authored-By: Claude <[email protected]> * docs(fork): rewrite FORK.md for the green-means-green gate TKT-336, final: rebased onto dev@441ad185 (Gemma's #10) confirmed feedback anomalyco#145 (v2.session.goal.get, mode=effect) is dead -- test:httpapi --mode effect went from 226/1 to 229/0, zero exceptions needed. Ethan's ruling: check reachability before building new skip machinery, and it paid off -- no per-mode scenario skip needed after all. FORK.md rewritten to match: the known-red-exceptions table becomes "Resolved CI reds" (history, not a live lookup), the milestone-1 gate section becomes "the merge gate: all green, any red blocks" per the operator's directive, and a new "Blacksmith runner sweep" section records every workflow's retarget/disable/if:false/untouched disposition with rationale, matching the pattern the ticket asked for (comment + un-skip condition at every change). Co-Authored-By: Claude <[email protected]> * ci(fork): skip e2e (windows) too -- systemic Playwright failures on first run e2e (windows) also never ran before TKT-336 (queued forever on blacksmith). Retargeting it to windows-latest was its first real execution, and it surfaced ~40 distinct e2e/regression/*.spec.ts failures ("element(s) not found" / toBeVisible timeouts), consistent across Playwright's own automatic retries. e2e (linux), the identical suite on ubuntu-latest in the same run, passed cleanly -- Windows- specific, and the volume/consistency reads as systemic rather than 40 independent app bugs. Same boundary as the Ripgrep-on-Windows unit finding: real E2E- environment debugging, out of scope for a CI-hygiene ticket. Filed as feedback anomalyco#156 with the run's artifact link (screenshots + traces). "Run app e2e tests" is now Linux-only; e2e (linux) stays in the gate, e2e (windows) reports green without asserting anything until anomalyco#156 is root-caused. FORK.md updated: new row in "Resolved CI reds", gate section corrected to say e2e (linux) only, not both platforms. Co-Authored-By: Claude <[email protected]> * ci(fork): move e2e out of the gate entirely -- corrected reading, symmetric Ethan's ruling on the corrected e2e data (feedback anomalyco#156): the "~40 systemic Windows failures" reading was wrong -- a grep artifact counting Playwright's test-discovery listing, not real failures. Real count: 2 failed/87 passed on Windows, then 1 failed/96 passed on a later Linux run (previously clean twice), different spec each time. Ordinary, low-rate E2E flakiness on both platforms, pre-existing and never observed before (e2e never ran on either platform pre-TKT-336). Un-skips the Windows e2e execution step (symmetry -- both platforms run, both observational, or the asymmetry itself misleads the next reader) and moves e2e out of the merge gate on BOTH platforms rather than reintroducing a Windows-only exception. The gate is now typecheck + unit(both) + test:httpapi. e2e stays running (not disabled) with a deterministic reading rule in FORK.md, so "observational" doesn't become "advisory and ignored" -- the exact trap this ticket exists to close: the SAME spec failing 3 consecutive runs is a real regression, filed no-judgment-required; scattered single-spec flakes across different specs are not. Stabilizing genuine Playwright flakiness belongs to the milestone-3 real-browser-matrix slice, not a CI-hygiene ticket. Co-Authored-By: Claude <[email protected]> --------- Co-authored-by: Felicity <[email protected]> Co-authored-by: Claude <[email protected]>
1 parent 6ff63c5 commit 2eeb091

20 files changed

Lines changed: 292 additions & 146 deletions

‎.github/workflows/beta.yml‎

Lines changed: 5 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,12 @@
11
name: beta
22

3+
# Fork note (TKT-336): schedule trigger disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
4+
# which queues forever on this org -- HOURLY cron made this the single largest
5+
# contributor to the accumulated queued-run backlog (24/day since Actions enablement).
6+
# Beta-channel sync is not meaningful on a fork that isn't publishing releases (see
7+
# publish.yml). workflow_dispatch kept as a manual escape hatch.
38
on:
49
workflow_dispatch:
5-
schedule:
6-
- cron: "0 * * * *"
710

811
jobs:
912
sync:

‎.github/workflows/containers.yml‎

Lines changed: 6 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1,13 +1,12 @@
11
name: containers
22

3+
# Fork note (TKT-336): push trigger disabled. Real container publish pipeline
4+
# (packages: write, pushes to ghcr.io/${{ github.repository_owner }} -- i.e. this org's
5+
# own registry namespace). Same safety reasoning as publish.yml: an internal
6+
# architecture-redesign fork should not auto-publish container images under the org
7+
# name. package.json is a common change path, so left enabled this would fire often.
8+
# workflow_dispatch kept as a manual escape hatch.
39
on:
4-
push:
5-
branches:
6-
- dev
7-
paths:
8-
- packages/containers/**
9-
- .github/workflows/containers.yml
10-
- package.json
1110
workflow_dispatch:
1211

1312
permissions:

‎.github/workflows/docs-locale-sync.yml‎

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,12 @@
11
name: docs-locale-sync
22

3+
# Fork note (TKT-336): push trigger disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
4+
# which queues forever on this org. Docs-locale-sync is not this fork's concern (not the
5+
# canonical docs source). workflow_dispatch added as a manual escape hatch. Un-disable
6+
# condition: retarget runs-on AND restore the push trigger below, if this fork ever
7+
# needs its own docs-locale sync.
38
on:
4-
push:
5-
branches:
6-
- dev
7-
paths:
8-
- packages/web/src/content/docs/*.mdx
9+
workflow_dispatch:
910

1011
jobs:
1112
sync-locales:

‎.github/workflows/docs-update.yml‎

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,12 @@
11
name: docs-update
22

3+
# Fork note (TKT-336): schedule trigger disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
4+
# which queues forever on this org -- every-12h cron meant it accumulated a fresh
5+
# unstartable run twice a day regardless of any push/PR activity. Docs-sync automation
6+
# is not this fork's concern (not the canonical docs source). workflow_dispatch kept as
7+
# a manual escape hatch. Un-disable condition: retarget runs-on AND restore the schedule
8+
# trigger below, if this fork ever needs its own docs-sync cadence.
39
on:
4-
schedule:
5-
- cron: "0 */12 * * *"
610
workflow_dispatch:
711

812
env:

‎.github/workflows/generate.yml‎

Lines changed: 9 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,15 @@
11
name: generate
22

3+
# Fork note (TKT-336): auto-trigger disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
4+
# which queues forever on this org and was accumulating unstartable runs on every push
5+
# to dev. Also a deliberate safety choice, not just a runner fix: this job auto-commits
6+
# and pushes generated output back to whatever branch triggered it, unreviewed -- this
7+
# fork's own convention (FORK.md) is that generated code is regenerated and committed
8+
# manually as part of a reviewed PR, never pushed unattended to dev. workflow_dispatch
9+
# kept as a manual escape hatch. Un-disable condition: retarget runs-on AND restore the
10+
# push trigger below, only if the unattended-push behavior is deliberately wanted here.
311
on:
4-
push:
5-
branches:
6-
- dev
12+
workflow_dispatch:
713

814
jobs:
915
generate:

‎.github/workflows/nix-eval.yml‎

Lines changed: 7 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,13 @@
11
name: nix-eval
22

3+
# Fork note (TKT-336): auto-triggers disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
4+
# which queues forever on this org (no Blacksmith app installed) and was accumulating
5+
# unstartable runs on every push/PR. Nix packaging validity is an upstream distribution
6+
# concern this fork does not maintain (no nix build/publish happens here). workflow_dispatch
7+
# kept as a manual escape hatch. See FORK.md. Un-disable condition: retarget runs-on to a
8+
# GitHub-hosted runner AND restore the push/pull_request triggers below, if this fork ever
9+
# needs to keep its own Nix packaging green.
310
on:
4-
push:
5-
branches: [dev]
6-
pull_request:
7-
branches: [dev]
811
workflow_dispatch:
912

1013
concurrency:

‎.github/workflows/nix-hashes.yml‎

Lines changed: 5 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -3,19 +3,13 @@ name: nix-hashes
33
permissions:
44
contents: write
55

6+
# Fork note (TKT-336): push trigger disabled. runs-on is blacksmith-4vcpu-ubuntu-2404,
7+
# which queues forever on this org, and the path filter includes package.json/bun.lock
8+
# -- changed often enough to be a real backlog contributor. Nix packaging is not this
9+
# fork's concern, same reasoning as nix-eval.yml. workflow_dispatch kept as a manual
10+
# escape hatch.
611
on:
712
workflow_dispatch:
8-
push:
9-
branches: [dev, beta]
10-
paths:
11-
- "bun.lock"
12-
- "package.json"
13-
- "packages/*/package.json"
14-
- "flake.lock"
15-
- "nix/node_modules.nix"
16-
- "nix/scripts/**"
17-
- "patches/**"
18-
- ".github/workflows/nix-hashes.yml"
1913

2014
concurrency:
2115
group: ${{ github.workflow }}-${{ github.ref }}

‎.github/workflows/notify-discord.yml‎

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,7 +6,12 @@ on:
66

77
jobs:
88
notify:
9-
runs-on: blacksmith-4vcpu-ubuntu-2404
9+
# Fork note (TKT-336): retargeted from blacksmith-4vcpu-ubuntu-2404. Low priority --
10+
# only fires on an actual GitHub release being published, which requires a deliberate
11+
# human act and won't happen via automation now that publish.yml's push trigger is
12+
# disabled -- but retargeting is free and keeps this consistent if a release is ever
13+
# cut by hand. Whether DISCORD_WEBHOOK is configured on this repo is unverified.
14+
runs-on: ubuntu-latest
1015
steps:
1116
- name: Send nicely-formatted embed to Discord
1217
uses: SethCohen/github-releases-to-discord@24d166886aee4646d448c8a389ff9e1ebcab3682 # v1.20.0

‎.github/workflows/opencode.yml‎

Lines changed: 10 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -8,11 +8,19 @@ on:
88

99
jobs:
1010
opencode:
11+
# Fork note (TKT-336): disabled -- same double-blocker as review.yml's /review
12+
# trigger (see FORK.md): runs-on is blacksmith-4vcpu-ubuntu-2404 (queues forever)
13+
# AND the step needs OPENCODE_API_KEY, not configured on this repo. Fires on any
14+
# PR/issue comment containing "/oc" or "/opencode", so left enabled it could add a
15+
# permanently-queued run at any time. `false &&` is a one-line, trivially revertible
16+
# off-switch that preserves the real condition beneath it. Un-disable condition:
17+
# both a GitHub-hosted runs-on AND an operator-provisioned OPENCODE_API_KEY secret.
1118
if: |
12-
contains(github.event.comment.body, ' /oc') ||
19+
false &&
20+
(contains(github.event.comment.body, ' /oc') ||
1321
startsWith(github.event.comment.body, '/oc') ||
1422
contains(github.event.comment.body, ' /opencode') ||
15-
startsWith(github.event.comment.body, '/opencode')
23+
startsWith(github.event.comment.body, '/opencode'))
1624
runs-on: blacksmith-4vcpu-ubuntu-2404
1725
permissions:
1826
id-token: write

‎.github/workflows/pr-management.yml‎

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,16 @@ on:
66

77
jobs:
88
check-duplicates:
9+
# Fork note (TKT-336): disabled, not just runner-blocked. Two independent problems,
10+
# same as review.yml's /review trigger (see FORK.md, "Does automated review run on
11+
# fork PRs? -- No"): runs-on is blacksmith-4vcpu-ubuntu-2404 (queues forever, no
12+
# Blacksmith app on this org), AND the step below needs OPENCODE_API_KEY, which is
13+
# not configured on this repo -- retargeting the runner alone would just turn a
14+
# silent queue into a visible failure on every PR. Fires on every PR open
15+
# (pull_request_target: [opened]), so left running would violate "every PR's checks
16+
# complete green" regardless of which half is broken. Un-disable condition: both a
17+
# GitHub-hosted runs-on AND an operator-provisioned OPENCODE_API_KEY secret.
18+
if: false
919
runs-on: blacksmith-4vcpu-ubuntu-2404
1020
permissions:
1121
contents: read

0 commit comments

Comments
 (0)