NEC-375: Renovate failures - investigate #339

Merged
nectenda-agent merged 7 commits from worktree-nec-375-renovate-failures-investigate into main 2026-11-07 02:57:42 +00:00
Collaborator

Task: NEC-375 — https://projectron.nerchure.com/tasks/397

Renovate itself has been healthy on every run; what failed were the checks on its pull requests. (A) The release-note job passed the pull request body through the environment, and Linux refuses one environment string of 128 KiB or more, so on Renovate's 250 KB bodies (#320, #337) the step died at exec with 'argument list too long' before node started, and the NEC-353 hold never ran either. Both scripts now read the body from the runner's event file (GITHUB_EVENT_PATH) and exit 2 when they cannot see it; a test spawns them on a 300,000-byte body, and a scan refuses any github.event reference in .forgejo that is not a named short scalar. (B) Only development dependencies carried a '## Changelog NONE', so Renovate's own image bump (#320) waited for a note nobody could keep in a body Renovate rewrites: action refs, the Renovate image and the Obsidian e2e pin now carry NONE, and scripts/renovate-config.test.mjs refuses NONE on any rule that can match something shipped. (C) Development majors were a batch of their own; #337 held fourteen and TypeScript 7 dropping baseUrl kept every one out. They now arrive one pull request each (monorepo families together), under renovate/dev-, never automerged. Review with Renovate's own rule matching found that without that prefix a package that is dev in one workspace and production in another (ws, yjs) shared a branch and the production major inherited NONE; fixed and tested. Also: the 'dependencies' and 'security' labels renovate.json asks for did not exist and were created; the TypeScript 7 migration is filed as NEC-376; and as a ride-along NEC-377, e2e.yml's push trigger now excludes renovate/**, since Renovate's rebases made main's own edits to its paths count as the branch's (three runs on 26 Oct). Verified by a probe pair: the same head started e2e.yml on this branch (run 1664) and not on renovate/nec-375-probe. Not verifiable until Renovate's next run on Monday 9 Nov: the notes Renovate actually writes, #337 autoclosing, #320 gaining its note.

Spec: docs/changes/NEC-375-renovate-failures/spec.md

Also fixes NEC-377: e2e.yml runs on Renovate rebases despite its paths filter

Changelog

NONE

Task: NEC-375 — https://projectron.nerchure.com/tasks/397 Renovate itself has been healthy on every run; what failed were the checks on its pull requests. (A) The release-note job passed the pull request body through the environment, and Linux refuses one environment string of 128 KiB or more, so on Renovate's 250 KB bodies (#320, #337) the step died at exec with 'argument list too long' before node started, and the NEC-353 hold never ran either. Both scripts now read the body from the runner's event file (GITHUB_EVENT_PATH) and exit 2 when they cannot see it; a test spawns them on a 300,000-byte body, and a scan refuses any github.event reference in .forgejo that is not a named short scalar. (B) Only development dependencies carried a '## Changelog NONE', so Renovate's own image bump (#320) waited for a note nobody could keep in a body Renovate rewrites: action refs, the Renovate image and the Obsidian e2e pin now carry NONE, and scripts/renovate-config.test.mjs refuses NONE on any rule that can match something shipped. (C) Development majors were a batch of their own; #337 held fourteen and TypeScript 7 dropping baseUrl kept every one out. They now arrive one pull request each (monorepo families together), under renovate/dev-, never automerged. Review with Renovate's own rule matching found that without that prefix a package that is dev in one workspace and production in another (ws, yjs) shared a branch and the production major inherited NONE; fixed and tested. Also: the 'dependencies' and 'security' labels renovate.json asks for did not exist and were created; the TypeScript 7 migration is filed as NEC-376; and as a ride-along NEC-377, e2e.yml's push trigger now excludes renovate/**, since Renovate's rebases made main's own edits to its paths count as the branch's (three runs on 26 Oct). Verified by a probe pair: the same head started e2e.yml on this branch (run 1664) and not on renovate/nec-375-probe. Not verifiable until Renovate's next run on Monday 9 Nov: the notes Renovate actually writes, #337 autoclosing, #320 gaining its note. Spec: `docs/changes/NEC-375-renovate-failures/spec.md` Also fixes NEC-377: e2e.yml runs on Renovate rebases despite its paths filter ## Changelog NONE
The release-note job passed the body to its scripts in the environment, and
Linux refuses any one environment string of 128 KiB or more: the step failed
at exec, "argument list too long", before node started. Renovate's bodies
carry every upstream changelog, and #320 (253 KB) and #337 (287 KB) could not
be judged at all, nor rescued by editing, which is how NEC-353's hold is
answered. Both scripts now read the body from GITHUB_EVENT_PATH, the payload
the expressions came from, and exit 2 when they cannot see it.

Task: NEC-375

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
The development batch took majors too, and Renovate split them into a batch
of their own: #337 held fourteen, and TypeScript 7 dropping baseUrl kept all
fourteen out. Majors now arrive one pull request each (monorepo families
still together), never automerged, still NONE.

Only development dependencies carried a release note, so Renovate's own image
bump (#320) waited on a person to write NONE into a body Renovate rewrites.
Action refs, Renovate's image and the Obsidian e2e pin now carry it, and
scripts/renovate-config.test.mjs refuses NONE on any rule that can match
something that ships.

Task: NEC-375

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Task: NEC-375

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Renovate rebases weeks-old branches onto main and force-pushes them, and the
paths filter on e2e.yml's push trigger compared old tip with new, so main's
own edits to e2e.yml counted as the branch's: three Renovate pushes on
26 October started two Obsidian jobs each. ci.yml's e2e covers those pull
requests through the same composite action.

Task: NEC-375
Also-fixes: NEC-377

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Task: NEC-375

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Review of NEC-375 ran the rules through Renovate's own matching: a package
that is a development dependency in one workspace and a production one in
another (ws, yjs) put both majors on one branch, named for the package, and
Renovate adds up the notes of every upgrade on a branch. The production bump
would have passed the release-note check with NONE. A dev- prefix keeps them
apart. The batch also lists every non-major type, so a pin no longer falls
out of it, and the .forgejo scan now refuses any event reference that is not
a named scalar, not only one ending in body.

Task: NEC-375

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Bump the e2e marker, to show the push trigger still fires off Renovate's branches
All checks were successful
Release note / release-note (pull_request) Successful in 21s
CI / build (pull_request) Successful in 5m56s
CI / e2e (pull_request) Successful in 15m27s
CI / promote (pull_request) Has been skipped
CI / build (push) Successful in 7m0s
e2e / multi (push) Successful in 15m25s
e2e / multi-newest (push) Successful in 15m28s
CI / e2e (push) Successful in 16m13s
CI / promote (push) Successful in 34s
61ed0aa973
The same head goes to renovate/nec-375-probe, where it should not.

Task: NEC-375
Also-fixes: NEC-377

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EV8ztaVFahJDbCBjrTQEBW
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
Nectenda/nectenda!339
No description provided.