test: lockingSpec advisory-lock + transaction tests need engine guards for H2 and SQL Server

[bpamiri]

Describe the bug

wheels.tests.specs.model.lockingSpec exercises advisory-lock and explicit-transaction behavior, but assumes the underlying DB supports these primitives. On engines/DBs where the primitive isn't supported, the spec errors instead of skipping gracefully:

• H2: H2 does not support advisory locks. (4 errors)
• SQL Server: The statement or function must be executed in the context of a user transaction. (4 errors)
• BoxLang on most DBs: Expected [false] to be true (3 errors — same suite, releases lock even when callback throws an exception)
• Oracle: Oracle advisory locks require DBMS_LOCK package setup. (4 — see ci: Oracle datasource broken across all engines — Invalid URL, missing DBMS_LOCK, constraint cleanup #2663)

Evidence

Compat-matrix run #25837380625 (2026-05-13). Lucee 6 + H2 = 4; Lucee 6 + SQL Server = 4; BoxLang across mysql/postgres/sqlite = 3.

Expected behavior

When the underlying DB doesn't support the locking primitive the spec is testing, the spec should be skipped for that DB (not errored). When it IS supported, the spec should pass.

The current state is "exercises feature, errors when feature is unavailable" — which makes the matrix unable to distinguish "lock implementation broken" from "lock not applicable to this DB."

Suggested fix shape

Two parts:

1. Guard the lock-feature specs so they call $skip() (or equivalent) when the current DB adapter reports no advisory-lock support. The Wheels DB adapters know which features they support; check that.
2. Investigate the BoxLang-on-supported-DBs Expected [false] to be true case separately — that suggests a real bug in lock-release-on-exception on BoxLang specifically. It's not a "feature unsupported" case (the DBs in question — mysql, postgres, sqlite — do support advisory locks), so it's likely a try/catch flow issue in the BoxLang adapter's locking path.

Additional context

• Part of the Core test suite has high failure/error rate on BoxLang (35 failures, 125 errors, 3526 passing) #2649 decomposition (Tier 3).
• Pairs naturally with ci: Oracle datasource broken across all engines — Invalid URL, missing DBMS_LOCK, constraint cleanup #2663 (Oracle's DBMS_LOCK setup) — same suite, partly the same fix shape.
• The BoxLang-specific lock release on exception failure may warrant its own ticket once investigated; for now it's lumped here because the spec is the same.

Related: #2649, #2663

[wheels-bot]

Wheels Bot — Triage

Classified as bug.

Suspected layer

model — see .ai/wheels/models/ for canonical patterns. The locking primitive lives in vendor/wheels/model/locking.cfc and the affected spec is vendor/wheels/tests/specs/model/lockingSpec.cfc.

Fix sketch

Two-part fix as the reporter outlines. First, add per-it (or per-describe) skip guards in vendor/wheels/tests/specs/model/lockingSpec.cfc that consult the current adapter advisory-lock / explicit-transaction support and call $skip() (or the WheelsTest equivalent) when the primitive is not available — the suite then reports as skipped on H2 / SQL Server / Oracle instead of erroring. The capability check likely needs a small accessor on the DB adapter (or a helper in vendor/wheels/model/locking.cfc) so the spec can ask "does this adapter support advisory locks?" rather than hard-coding engine names. Second, the BoxLang Expected [false] to be true failure on mysql/postgres/sqlite is a real bug in lock-release-on-exception on BoxLang and should be carved out into a separate issue once reproduced — propose-fix should leave that case alone and only land the skip-guard work.

Confidence: medium

Cross-engine concern detected (Lucee vs Adobe vs BoxLang behavior differs across the affected DBs, and the BoxLang sub-case is a distinct real bug) — auto-downgrade from high to medium per the cross-engine rule.

<!-- wheels-bot:triage:2665 -->
<!-- wheels-bot:triage-class:bug -->
<!-- wheels-bot:triage-confidence:medium -->

[wheels-bot]

Wheels Bot — Fix proposed

Draft PR: #2670

Spec: vendor/wheels/tests/specs/model/lockingSpec.cfc — failing → passing
Implementation:

• vendor/wheels/databaseAdapters/Base.cfc — new $supportsAdvisoryLocks() capability (default false)
• vendor/wheels/databaseAdapters/PostgreSQL/PostgreSQLModel.cfc — override true
• vendor/wheels/databaseAdapters/MySQL/MySQLModel.cfc — override true
• vendor/wheels/databaseAdapters/SQLite/SQLiteModel.cfc — override true (no-op but contract-honoring)
• vendor/wheels/databaseAdapters/MicrosoftSQLServer/MicrosoftSQLServerModel.cfc — override false (sp_getapplock requires an enclosing transaction; carved out for separate work)

H2 / Oracle / CockroachDB inherit Base's false. The BoxLang release-on-exception failure on supported DBs (mysql/postgres/sqlite) is explicitly left out of scope per the triage.

A human review is required before merge. Reviewer A and Reviewer B will weigh in shortly. Doc updates (MDX guides, .ai/wheels/, CLAUDE.md) are handled separately by bot-update-docs.yml.