Skip to content

Report runtime boundary increases a PR has already accepted as notices - #15183

Merged
cderv merged 1 commit into
mainfrom
fix/runtime-boundary-accepted-baseline
Oct 23, 2026
Merged

cderv merged 1 commit into
mainfrom
fix/runtime-boundary-accepted-baseline

Conversation

@cderv

@cderv cderv commented Oct 23, 2026

Copy link
Copy Markdown
Member

Follow-up to #15153.
In pull request mode the runtime boundary check compares the merge commit with the base branch only, so it never sees the PR's own tools/runtime-boundary-baseline.json.
#15180 ran the maintainer step (--update-baseline --accept-increase, tests 832 → 838) and the check still reported the six new calls as warnings saying "a maintainer accepts the new count before merging", with the "Maintainer action before merging" block in the job summary (run).
A reviewer cannot tell from the check whether the step is done.

Fix

PR mode now also reads the baseline in the merge commit.
When the PR raised a count there and the merge result does not exceed it, each finding for that count is a notice ("accepted by this PR's tools/runtime-boundary-baseline.json") on the added line, and the summary lists it under "Accepted in this PR" with the old and new baseline value.
This is the comparison the push check on main makes after merge.

A count is accepted only when the PR raised it.
Room left in main's baseline (a count below its baseline because nobody lowered it) does not hide a new call: it stays a warning.
A baseline raised before updating the branch with main, and now too low for the merge result, also keeps the warnings and the maintainer block.
Errors for a Deno.* API that src/deno_ral replaces do not depend on the baseline and are unchanged.

The new test runs the script with quarto run --dev ... --against HEAD in a temporary git repo, like CI does: before the fix it showed the same warning and maintainer block as #15180.

Checklist

I have (if applicable):

  • referenced the GitHub issue this PR closes
  • updated the appropriate changelog in the PR
  • ensured the present test suite passes
  • added new tests
  • created a separate documentation PR in Quarto's website repo and linked it to this PR
AI-assisted PR
  • AI tool used: Claude Code (Opus 5.5)
  • Codebase grounding: local clone
  • Human review: I have reviewed, tested, and verified the AI-generated content before submitting.

Note: autonomous AI agents submitting PRs without human oversight are not permitted — see the Code of Conduct.

In pull request mode the check compared the merge result only with the
base branch, so a PR that had already run --update-baseline
--accept-increase still got one warning per new call and the
"Maintainer action before merging" block (seen on PR 15180, which
raised tests from 832 to 838). Reviewers could not tell whether the
step was still pending.

The check now also reads the working tree's baseline. A count the PR
raised there and does not exceed is reported as a notice plus an
"Accepted in this PR" line; that is the state the push check on main
will see after merge. Slack already present in the base baseline does
not count as acceptance, so new calls still get a warning unless the
PR raises the number itself. A baseline raised before updating from
main, and now too low, keeps the warnings. Errors for APIs that
src/deno_ral replaces are unchanged.
@posit-snyk-bot

posit-snyk-bot commented Oct 23, 2026 •

Copy link
Copy Markdown
Collaborator

✅ Snyk checks have passed. No issues have been found so far.

Status Scan Engine Critical High Medium Low Total (0)
✅ Open Source Security 0 0 0 0 0 issues
✅ Licenses 0 0 0 0 0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

@cderv
cderv merged commit 3a689f2 into main Oct 23, 2026
50 of 52 checks passed
@cderv
cderv deleted the fix/runtime-boundary-accepted-baseline branch October 23, 2026 13:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants