# Leaner, steadier PHPUnit runs for upcoming releases > [!info] Ported from an automated deep-read — not Willett's words except inside `> [!quote]` blocks. > Sourced 2026-08-16 from Make WordPress Core. > [!question] Lingering Questions > - Which specific database combinations were removed, and what evidence established they were actually redundant rather than merely unexercised in the sample window? > - How many comparable CI runs support the reported 52%/54% percentages, and what are the confidence intervals? > - Will a scheduled full-matrix "sentinel" run guard against the blind spots a trimmed matrix could hide? > [!tip] Article Ideas > - A one-page "matrix coverage decision record" template — what was removed, what evidence justified it, what would trigger restoring it — generalizable to any project trimming a CI test matrix. > [!link] Cross Reference > - [[3. Resources/Permanent Notes/Maintenance Visibility Functions as Funding Infrastructure]] — same genre of move: making previously-invisible infrastructure work (here, CI capacity) visible and reportable > - [[3. Resources/Permanent Notes/Prevented Work Is Invisible to Velocity Metrics]] — the CI-reliability half of this post (bounded Docker retries, shared Gutenberg artifact) prevents failures that a rerun-rate metric barely captures and a raw pass/fail count wouldn't show at all > - [[3. Resources/MOCs/Open Source Governance MOC]] > [!quote] Vocabulary > - *boundary matrix* — testing every supported PHP version but only the database pairings that sit at meaningful compatibility edges, instead of every PHP × database combination > - *rerun-to-green* — the fraction of CI runs that fail at least once and only pass after a rerun; a proxy for infrastructure flakiness distinct from actual test failures ## Bib - **Source**: Lance Willett, ["Leaner, steadier PHPUnit runs for upcoming releases"](https://make.wordpress.org/core/2026/07/30/leaner-steadier-phpunit-runs-for-upcoming-releases/), Make WordPress Core, 2026-07-30 - **Reader doc**: [[3. Resources/Readwise/Articles/Leaner, steadier PHPUnit runs for upcoming releases]] ## Author's thesis WordPress Core trimmed its GitHub Actions PHPUnit matrix before 7.1 by keeping full PHP-version boundary coverage while dropping redundant database combinations, and paired that with two reliability fixes (a single shared Gutenberg-build fetch per workflow run, bounded Docker pull retries). The measured result is about 52% fewer jobs and 54% fewer job-minutes per run, with rerun-to-green frequency falling from roughly 68% to 36% — capacity and stability gains that say nothing yet about individual test speed or defect detection, which the post explicitly leaves as separate, unsolved problems. ## Key arguments (≤3) 1. **Boundary matrices preserve PHP coverage while cutting redundant database pairings** — the coverage contract (every supported PHP version) stays intact; what's removed is duplicated work, not compatibility guarantees. 2. **Matrix trimming roughly halves CI capacity use** — ~52% fewer jobs, ~54% fewer job-minutes per run. 3. **Centralizing a shared, verified dependency reduces infrastructure-caused reruns** — fetching Gutenberg once per run (with retained checksum/size verification) and bounding Docker pull retries moved rerun-to-green frequency from ~68% to ~36%. ## Claim atoms ### Boundary matrices preserve PHP coverage > [!quote] ID:rw_01kz91d6x3kq51677bc6b0bwp8 > Trimmed the PHPUnit matrix to boundary PHP versions (#12719 on trunk, #12720 for 7.0). Full PHP coverage kept, redundant database combinations dropped. Where/why: matrix-trim section — frames the change as removing redundant combinations while retaining an explicit compatibility boundary, not weakening the contract. ### Matrix trimming halves CI work > [!quote] ID:rw_01kz91d7m1110kcgrccz73r53s > Per run: ~52% fewer jobs and ~54% fewer job-minutes. Where/why: results section — job count and job-minutes quantify capacity reclaimed per workflow run. ### Reliability fixes reduce rerun waste > [!quote] ID:rw_01kz91d5znhnmw2746f272qwj0 > Runs needing a rerun to go green roughly halved, from ~68% to ~36%. Where/why: reliability-results section — rerun frequency exposes infrastructure instability that a simple pass/fail status hides. ## Permanent candidates - **"Remove demonstrably redundant test combinations before weakening an explicit compatibility boundary"** — the strongest generalizable claim: efficiency work here targets duplication, not the coverage contract itself. Promoted below. ## Processing checklist - [x] Connect into the relevant MOC — [[3. Resources/MOCs/Open Source Governance MOC]] - [x] Create Permanent note for the redundant-combinations claim - [ ] Reproduce the before/after GitHub Actions metrics with published sample sizes before citing the exact percentages elsewhere ## Links - [[3. Resources/MOCs/Open Source Governance MOC]] — CI capacity and coverage-contract decisions as release-engineering governance - [[3. Resources/Permanent Notes/Cut Redundant CI Combinations Before Weakening a Coverage Contract]] — the new Permanent note this source generated