Add corpus regeneration diff CI

[sirreal]

Implements the corpus-diff check that the #262 and #270 review rounds relied on manually: parse a pinned corpus of WordPress core source (wp-includes from the WordPress 6.8 tag) with the parser at the PR's merge base and at its head, normalize both JSON outputs with prep-diff.php, and diff. Every hunk in the diff is a behavior change the PR makes — intended and explained, or a regression. This check has repeatedly caught regressions no unit test covered.

Decisions taken (also documented in the workflow header):

• Non-blocking to start. The job always succeeds; the diff is uploaded as an artifact and the job summary reports the hunk/line count (or 0 hunks). Make it blocking with an escape hatch (label or committed expected diff) once the signal has proven trustworthy.
• The head checkout's tooling drives both sides. tools/export-corpus.php takes the parser root as a parameter, so one copy exports through the base checkout and the head checkout; prep-diff.php from head normalizes both. Tooling changes therefore never masquerade as parser changes; a change to prep-diff.php itself shows its normalization effect in the diff and is reviewed like any other change.
• PHP 7.4 — the supported floor, matching the oldest runtime in the unit-test matrix.
• Corpus scope: wp-includes only for now (~1000 files, a few minutes to parse). wp-admin/includes can be added later for hook/uses coverage at the cost of runtime.
• Separate workflow from unit tests: different failure semantics.

Determinism: pinned corpus tag (cached by tag), sorted file list via the existing get_wp_files(), LC_ALL=C, the same PHP binary for both sides, JSON_PRETTY_PRINT on both. The export script routes PHP diagnostics to stderr so stdout stays pure JSON (vendored code emits deprecation notices on newer PHP).

Smoke-tested locally: exporting tests/source through tools/export-corpus.php and piping through prep-diff.php produces clean normalized JSON.

Design doc and manual procedure this implements were written during the #262/#270 review work.

Update

Corpus pinned to WordPress 7.0.4. 7.0.4 is the latest WordPress release; pinning the point release rather than the 7.0 major tag tracks the patches that land in wp-includes, so the corpus matches what is shipped. WordPress 7.0 also raised core's minimum PHP to 7.4, exactly this job's PHP floor. Verified locally against wp-includes@7.0.4: 1039 PHP files, all present in the export, no parse errors, ~49 MB of JSON — both of the job's guards (>= 500 files, >= 1 MB of JSON) still hold.

Manual procedure documented in the README. A new "Corpus diff" section records what the check is, how to run it locally with the repo's own tools/export-corpus.php and prep-diff.php (merge-base worktree, export both sides over the same corpus, normalize, diff), and the tooling policy the workflow header already states: the head checkout's tooling drives both sides.

🤖 Generated with Claude Code