test(model): guard lockingSpec advisory-lock tests on adapters without standalone support

[wheels-bot]

Summary

Adds a $supportsAdvisoryLocks() capability accessor to the database adapter base + overrides on PostgreSQL/MySQL/SQLite (true) and MicrosoftSQLServer (false, pending an implicit-transaction wrapper). lockingSpec consults the new capability via a beforeEach { skip(...) } so advisory-lock tests are now reported as Skipped (with a reason) on H2, SQL Server, Oracle, and CockroachDB instead of erroring. The CockroachDB silent return hack is removed in favor of the same skip path, so compat-matrix can distinguish "lock implementation broken" from "lock not applicable to this DB."

The BoxLang Expected [false] to be true failure on the release-on-exception case is intentionally left out of scope per the triage — that's a real bug on a DB that does support advisory locks (mysql/postgres/sqlite) and warrants its own ticket.

Related Issue

Fixes #2665

Type of Change

• Bug fix
• New feature
• Enhancement to existing feature
• Documentation update
• Refactoring

Feature Completeness Checklist

• DCO sign-off -- Every commit carries Signed-off-by: (use git commit -s); see CONTRIBUTING.md
• Tests -- Failing-then-passing capability-accessor spec; existing withAdvisoryLock describe now skips on unsupported adapters
• Framework Docs -- handled by bot-update-docs.yml follow-up
• AI Reference Docs -- handled by bot-update-docs.yml follow-up
• CLAUDE.md -- handled by bot-update-docs.yml follow-up
• CHANGELOG.md -- entry added under [Unreleased] → Changed
• Test runner passes -- Lucee 7 + SQLite via local runner: wheels.tests.specs.model reports 811 pass, 0 fail, 0 error, 11 skipped (all 11 lockingSpec specs pass on SQLite; pre-existing skips on raceconditionSpec / useIndex are unrelated)

Test Plan

• Local run: hit /wheels/core/tests?db=sqlite&format=json&directory=wheels.tests.specs.model after the CI setup-wheels-test-env action — confirms 0 errors and all lockingSpec specs passing
• Compat matrix should now show the previously-erroring tests as Skipped on H2, SQL Server, Oracle, CockroachDB (4 errors → 4 skips each)

[wheels-bot]

Wheels Bot — No doc updates

Reviewed this PR's diff and found no docs that need updating (purely internal test-fix: the new $supportsAdvisoryLocks() capability accessor is $-prefixed, no existing user guide or .ai/ reference covers withAdvisoryLock, and the CHANGELOG entry was already added in the PR).

[wheels-bot]

Wheels Bot — Reviewer A

TL;DR: This is a clean, well-scoped fix that replaces the ad-hoc CockroachDB return guards in lockingSpec with a proper capability-accessor pattern ($supportsAdvisoryLocks()) across all database adapters. The mechanics are correct, the test structure is sound, and cross-engine compatibility is fine. One minor nit about a redundant override, but nothing blocking.

---

Correctness

The change does exactly what the PR description says. The flow is:

1. Base.cfc provides a default $supportsAdvisoryLocks() → false (safe default — unknown adapters do not claim advisory-lock support).
2. PostgreSQL, MySQL, and SQLite override to true.
3. SQL Server overrides to false (same as the base, but with an explanatory comment).
4. Oracle, H2, and CockroachDB inherit false without an explicit override — correct, since all three throw Wheels.AdvisoryLockNotSupported from their lock methods (verified in vendor/wheels/databaseAdapters/{Oracle,H2,CockroachDB}/).
5. lockingSpec guards the withAdvisoryLock describe block with a beforeEach skip, and the old var _isCockroachDB + if (_isCockroachDB) return; pattern is fully removed.

All code paths are covered. The new capability-accessor describe block runs unconditionally and verifies that the method exists and returns a boolean on every adapter — a sensible smoke test.

---

Conventions

Minor nit — redundant SQL Server override

vendor/wheels/databaseAdapters/MicrosoftSQLServer/MicrosoftSQLServerModel.cfc (added lines):

public boolean function $supportsAdvisoryLocks() {
    return false;
}

This returns exactly what Base.cfc already returns. Oracle, H2, and CockroachDB (also non-supporting) correctly inherit false without an explicit override. Keeping an explicit false only for SQL Server creates a "why is this one different?" reader puzzle.

The comment explaining the implicit-transaction wrapper requirement is genuinely useful — but it could live in Base.cfc's method comment (perhaps under a ## Known non-supporting adapters list) rather than in a redundant override. Alternatively, a follow-up issue tracking the SQL Server implicit-transaction wrapper work is the right home for that detail.

This is a nit only — the code is not wrong, just inconsistent.

---

Cross-engine

No issues. Specifically checked:

• public access with $ prefix — all four adapter overrides and the base method use public boolean function $supportsAdvisoryLocks(). Per CLAUDE.md "Private mixin functions not integrated": mixin CFCs must use public + $ prefix. Database adapter CFCs are not mixin targets, but public is still correct here since the test spec calls the method directly.

• Closure accessing g in beforeEach — g = application.wo is an unscoped assignment inside run(), which places it in the CFC's variables scope (not local). Closures close over the declaring component's variables scope on Lucee, Adobe CF, and BoxLang — this is the same reason the existing it() closures can call g.model(...). No regression.

• skip() inside beforeEach closure — skip() is inherited from BaseSpec and resolves the same way expect() does in existing it() closures. The call is unqualified and resolves through the component's scope, consistent with every other WheelsTest BDD spec in the repo.

• No bracket-notation function calls, no client scope, no application scope members — none of the Adobe CF / BoxLang gotchas apply.

---

Tests

The new describe("Tests that withAdvisoryLock") beforeEach correctly guards all four it() blocks. The new describe("$supportsAdvisoryLocks adapter capability") has one unconditional test that fires on every adapter and checks the return type. The PR description reports 811 pass, 0 fail, 0 error, 11 skipped on Lucee 7 + SQLite, which is the expected outcome. Compat-matrix should now show "Skipped" (not error) on H2, SQL Server, Oracle, and CockroachDB runs. Good.

---

Docs

• CHANGELOG [Unreleased] → Changed entry present.
• PR body notes that .ai/ and CLAUDE.md doc updates are handled by the bot-update-docs.yml follow-up.

---

Commits

Single commit: test(model): guard lockingSpec advisory-lock tests on adapters without standalone support

Type test is in the allowlist. Scope model is appropriate. Subject is 75 characters, not ALL-CAPS. Passes commitlint. DCO sign-off flagged as verified in the PR checklist.

[wheels-bot]

Wheels Bot — Reviewer B (round 1)

A's review is accurate and well-reasoned on the main substance. One false positive in the SQL Server nit reasoning, one minor missed item on the BoxLang tracking issue, and one very minor CHANGELOG section placement question. None are blocking.

Sycophancy

None detected. A filed a COMMENTED review (not APPROVED), raised a concrete nit, and verified cross-engine concerns rather than rubber-stamping.

False positives

• SQL Server 'redundant override' nit is slightly mis-framed. A groups SQL Server with Oracle, H2, and CockroachDB as adapters that "correctly inherit false" without overrides. But those three throw Wheels.AdvisoryLockNotSupported from $acquireAdvisoryLock — they are categorically "no advisory lock concept here". SQL Server's $acquireAdvisoryLock is a real sp_getapplock implementation that succeeds inside a transaction but fails outside one. The explicit false + comment on SQL Server is not decorative inconsistency; it documents a specific, actionable gap ("needs an implicit transaction wrapper") that is architecturally distinct from the three non-supporting adapters. A's conclusion ("not wrong") is correct, but suggesting the comment belongs in Base.cfc under "Known non-supporting adapters" would muddy that distinction. The override should stay where it is.

Missed issues

• BoxLang advisory lock bug — no tracking issue verified. The PR body explicitly states the BoxLang Expected [false] to be true failure on the release-on-exception case "warrants its own ticket." A acknowledged it was left out of scope but did not verify that a dedicated issue was opened. Issue Core test suite has high failure/error rate on BoxLang (35 failures, 125 errors, 3526 passing) #2649 is a broad BoxLang failures umbrella; the advisory lock regression is a distinct, named failure mode. A reasonable reviewer check here is: is there an issue linked? If not, ask for one before merge so the fix is not silently dropped.

• CHANGELOG section placement (very minor). The entry sits under [Unreleased] -> Changed. This PR is a test-only fix with commit type test; conventional placement would be Fixed (compat-matrix was erroring, now it skips correctly). Not a blocking concern, but A did not flag it.

Verdict alignment

A's COMMENTED verdict is consistent with the finding of one non-blocking nit. I would reach the same verdict.

Convergence

Aligned on outcome: no changes required before merge, but requesting a BoxLang tracking issue linked from the PR body (or confirmation that #2649 is the designated home) is a reasonable ask before the PR moves out of draft. The SQL Server nit reasoning is slightly imprecise but the actionable conclusion is correct and does not need to be revisited. Joint recommendation: converged-approve pending confirmation of the BoxLang follow-up ticket.