Repository navigation
Conversation
Experiment for #442: turbocov in PCOV-impersonation mode, so the existing PCOV driver is exercised unchanged. Both jobs upload their clover report for comparison. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
Building the extension with PIE costs ~13 s per run. The cache key pins the PHP patch version and the turbocov release, so a new release or PHP update triggers a rebuild. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
Third baseline next to Xdebug and turbocov for #442, uploading its clover report as `clover-pcov`. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
… on the PR The job renders every source line where Xdebug, PCOV and turbocov disagree and keeps it in a single PR comment, found by a marker and rewritten on each push. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
Coverage driver diffEach row is one job: the same suite (
turbocov vs Xdebug
233 line(s) in 105 file(s) differ✅ executed, ❌ executable but never executed, ➖ not executable: absent from that driver's report.
turbocov vs PCOV
70 line(s) in 21 file(s) differ✅ executed, ❌ executable but never executed, ➖ not executable: absent from that driver's report.
|
…anges Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
ci: limit the experiment branch to PHP 8.4 and drop the 32-bit, pull_request and security runs turbocov is the subject, Xdebug and PCOV are baselines, each compared in one collapsible section. The trimmed CI keeps one regular test run next to the coverage jobs. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
Each baseline job runs the suite without coverage, with Xdebug or PCOV, then with turbocov, so timings share one machine. Every job rewrites only its own section of the PR comment; publishing is serialized through a concurrency group so the two read-modify-write cycles do not drop each other's section. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
PCOV also instrumented `tests/`, which inflated its timing against turbocov. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
The line-state legend moves into each diff spoiler, next to the marks it explains. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
…s report differently ci(coverage): run the driver comparison on PHP 8.5 The probe ports the turbocov fixtures for skipped operands, language constructs, edge cases and property hooks, plus asymmetric visibility, the pipe operator, clone with and a closure in a constant expression. Each function runs with one set of arguments, so lines one call skips stay visible in the per-line diff. PHP 8.5 is the newest version turbocov supports. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
…branch Changes to testo.php and the sources started mutation testing and the code-style fixer alongside the comparison. Assisted-By: Claude Opus 5.5 <noreply@anthropic.com>
🔍 What was changed
coverage-driver.yml, called once per baseline (Xdebug, PCOV). Each job runs the suite without coverage, with the baseline, then with turbocov on the same runner, and diffs the two clover reports.PcovDriveris used unchanged. The build is cached per PHP and turbocov version.Review notes
-d.pcovname.core:plugin:bridge). PCOV takes a single root, so its exclude pattern narrows it to the same set.Checklist