Skip to content

refactor: breakage-suite.yml decisions in tested scripts - #172

Merged
ocots merged 1 commit into
mainfrom
refactor/breakage-suite-shell
Nov 14, 2026
Merged

ocots merged 1 commit into
mainfrom
refactor/breakage-suite-shell

Conversation

@ocots

@ocots ocots commented Nov 13, 2026 •

Copy link
Copy Markdown
Member

Second step of #161 (after update-readme.yml, #167).

  • scripts/breakage/cell.jl: one cell — clone, pick the ref (latest = default branch, stable = newest v* tag, else the given ref), run the tests or the docs with the package under test developed, read the resolved version, write the cell file with TOML.print (the heredoc did not quote values).
  • scripts/breakage/pr.jl: facts (base, fork), baseline (the newest completed non-PR run on the base commit, its artifact), comment (update the comment that starts with the group's marker, or create it). Replaces the gh api shell and the github-script step; uses scripts/shared/github.jl and report.jl's marker.
  • report.jl --base-sha SHA: reads both versions with TOML instead of sed.
  • Inline shell of the workflow: ~102 → 24 lines, wiring only.

Behaviour changes, both deliberate:

  • A failed clone fails the cell's job (exit 1, no cell file; the report counts the cell as missing). Before, cd repo failed silently and the step went on in the wrong directory.
  • The downstream package is cloned into a temporary directory, not repo/ in the workspace.

Checked:

  • 144 breakage tests pass locally (34 report incl. --base-sha on a real git repo, 68 consumers, 22 cell, 20 pr); actionlint clean.
  • Against real GitHub: pr.jl facts and pr.jl baseline on CTBase.jl PR #653 (base found, the push run's breakage-packages artifact downloaded).
  • One real cell end to end: CTBase (local) under test, CTLie at stable → v0.3.1, tests pass, compat v0.30.13 (5 min).
  • Not run: the workflow itself. To test before merging: a CTBase test PR whose Breakage.yml calls breakage-suite.yml@refactor/breakage-suite-shell with ctactions_ref: refactor/breakage-suite-shell, then the label run breakage (ctactions_ref alone moves only the scripts, not the workflow file).

🤖 Generated with Claude Code

….jl)

The cell (which ref to test, the test or doc run, the cell file) moves to
scripts/breakage/cell.jl; the pull request facts, the baseline run on the base commit
and the comment kept by marker move to pr.jl, on scripts/shared/github.jl; report.jl
reads the two versions itself (--base-sha) instead of sed. Inline shell goes from about
102 lines to 24, all wiring. The cell file is written with TOML.print, so a value is
quoted correctly. A failed clone now fails the job instead of testing the wrong
directory. Refs #161.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ocots

ocots commented Nov 14, 2026

Copy link
Copy Markdown
Member Author

Tested in a real package: throwaway draft PR CTBase.jl#654 (Breakage.yml calling this branch with ctactions_ref, version bumped 0.30.13 → 0.30.14, run dispatched with pr=654 as the label does). Run https://github.com/control-toolbox/CTBase.jl/actions/runs/43574736731: all 16 cells ran, report and status-end succeeded, the PR comment was created, the commit status went to success. The release check read both versions (--base-sha), took the baseline of the push run on the base commit, and answered: OptimalControl and OptimalControlProblems already fail on the base (said nothing about the release), no downstream release broken by 0.30.13 → 0.30.14. PR closed, branch deleted.

@ocots
ocots merged commit d6072e8 into main Nov 14, 2026
4 checks passed
@ocots
ocots deleted the refactor/breakage-suite-shell branch November 14, 2026 11:41
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.

1 participant