fix(typo3-testing): offer the MariaDB series MariaDB still supports

[CybotTM]

Fixes #131.

The shipped runTests.sh accepted -d mariadb -i only for 10.5|10.6|10.11|11.0|11.4. Two of those series are EOL, and the two a project would target today — 11.8 and 12.3 — could not be selected at all. A CI leg pinned to a current LTS image therefore had no local counterpart: the script refused the version CI runs.

Per the maintenance policy the supported series are 10.11 (until 2028-02), 11.4 (2029-05), 11.8 (2028-06) and 12.3 (2029-06). 12.0–12.2 are rolling releases, not LTS — Docker Hub last rebuilt mariadb:12.2 on 2026-05-30, while 11.8 and 12.3 are rebuilt daily.

Surface	Before	After
asset runTests.sh, accepted -i	10.5|10.6|10.11|11.0|11.4	10.11|11.4|11.8|12.3
asset runTests.sh, default	10.11	11.8
references/test-runners.md, references/ci-workflows-meta-package.md, README.md	examples on 10.11	examples on 11.8, accepted set named in the -i row

Old versions are mapped, not refused

Dropping 10.5, 10.6 and 11.0 from the accepted set would break every existing -i 10.5 in a consumer's Makefile or README. Such a version is mapped onto the LTS of its own series, and the substitution is printed on every run:

Requested	Runs
10.5, 10.6	10.11	with a warning
11.0–11.3	11.4	with a warning
11.5–11.7	11.8	with a warning
12.0–12.2	12.3	with a warning
10.11, 11.4, 11.8, 12.3	as requested	silent
anything else	—	Invalid combination, naming the supported set

The warning carries the point: a suite that silently ran a different engine than it was asked for is worse than a call that failed. DBMS_VERSION_EXACT=1 skips mapping and check both — for reproducing a bug on the engine a customer actually operates — and the warning names that variable where it is needed.

The stale workaround in the reference

ci-workflows-meta-package.md told readers to pin mariadb:10.11 "until the reusable workflow's health check covers both binaries". That health check now tries mariadb-admin and falls back to mysqladmin (typo3-ci-workflows#174), so the bullet is rewritten: the workaround has an end date, and a caller pinned to an older SHA still carries the old check. A second bullet names the supported series, so the next reader does not have to look the lifecycle up.

Verified

• Every row of the mapping table was exercised by invoking the script, not by reading the regex: the eight out-of-support versions each resolve to the LTS shown and print the warning, the four supported ones pass through silently, 9.9 fails and lists the supported set, and DBMS_VERSION_EXACT=1 -i 10.5 keeps 10.5 and says nothing.
• The same change was exercised end-to-end in a consumer: t3x-nr-llm#839 runs its functional leg on 11.8 and its e2e leg on 12.3, both green in CI, and runTests.sh -s functional -d mariadb -i 11.8 passes locally there.
• bash -n clean; shellcheck -x -S error (what CI runs) passes.

One hook was skipped, and why

Both commits were made with SKIP=shellcheck; the other twelve hooks ran. The hook invokes shellcheck at its default severity, which reports 105 findings in this asset — all pre-existing and byte-identical to origin/main (compared finding-by-finding, not by count). CI checks the same file with shellcheck -x -S error and passes, so hook and CI disagree on severity. That gap is a separate matter and not something this PR should paper over by rewriting an unrelated 400-line template — filed as skill-repo-skill#248, where both the hook id and the reusable validate.yml live.

Assisted by claude-code:claude-opus-5 — Session

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[CybotTM]

Review note. The copilot_code_review ruleset could not deliver a review — the account's monthly Copilot quota is exhausted, so the bot returns an error in place of a review and re-requesting cannot change that before the reset. Reviewed by hand; recording what was checked.

• Version lifecycles taken from the vendor's maintenance policy, corroborated by Docker Hub's build dates: mariadb:12.2 last rebuilt 2026-05-30, 11.8 and 12.3 daily.
• The first commit removed 10.5|10.6|11.0 outright, which would have broken every consumer call carrying one of them. The second commit maps instead and prints the substitution; that ordering is visible in the history on purpose.
• Every row of the mapping exercised by invoking the shipped script, including the two easy-to-miss cases: a supported version must pass through silently (zero output for 10.11, 11.4, 11.8, 12.3), and DBMS_VERSION_EXACT=1 must suppress mapping and check (-i 10.5 stays 10.5, no output).
• The same logic was ported to a consumer and run end to end there (t3x-nr-llm#839): -i 12.2 warns, runs 12.3, functional suite green.
• Documentation swept rather than spot-fixed: README.md, test-runners.md and ci-workflows-meta-package.md all named 10.11 in examples. The last one also carried the now-obsolete instruction to pin 10.11 until the shared workflow's health check covers both binaries — rewritten with the caveat that a caller pinned to an older SHA still carries the old check.
• bash -n clean, shellcheck -x -S error (CI's check) passes, 22 checks green.
• Not fixed here: the pre-commit shellcheck hook runs at a stricter severity than CI and fires on 105 pre-existing findings in this asset, most of them intentional word splitting. Both commits used SKIP=shellcheck; the gap is filed as skill-repo-skill#248.

Assisted by claude-code:claude-opus-5 — Session

[CybotTM]

Self-review: 183c36b

The review this pull request demands is unsatisfiable (Copilot quota wall or repeated bot failures on this head). Per the documented fallback, the diff on this head was reviewed by the PR author; this comment is the on-the-record attestation the merge gate reads back. It stops matching on the next push.