ci: Oracle datasource broken across all engines — Invalid URL, missing DBMS_LOCK, constraint cleanup

[bpamiri]

Describe the bug

The Oracle database test path in the compat-matrix is broken at multiple layers, affecting every engine that exercises it (Lucee 6/7, Adobe 2023/2025, BoxLang). The failures fall into three sub-clusters:

1. Datasource registration failure (Lucee 6) — Oracle test JSON for Lucee 6 contains java.sql.SQLException: Invalid Oracle URL specified and reports 0 specs run. The connection never establishes.
2. DBMS_LOCK package not installed (any engine) — wheels.tests.specs.model.lockingSpec errors with Oracle advisory locks require DBMS_LOCK package setup. (4 occurrences).
3. Test cleanup leaves constraints behind — ORA-02264: name already used by an existing constraint and ORA-01003: no statement parsed (~6 occurrences in migratorSpec and bulkOperationsSpec). Same root cause class as the Lucee/MySQL Duplicate foreign key constraint name issue tracked separately.

Evidence

Compat-matrix run #25837380625 (2026-05-13). Oracle errors on BoxLang totaled 134 — the highest count per DB. On Lucee 6 the count was 0 because tests didn't run (datasource failure).

Expected behavior

Oracle tests should at minimum run to completion on every engine (currently fails on Lucee 6 before tests start). The advisory-lock and constraint-cleanup failures should be deterministic and either pass or skip gracefully.

Likely fix areas

This is CI-infrastructure work, not framework code. Three independent threads:

1. Lucee 6 + Oracle JDBC URL — Lucee 6's CFConfig.json for Oracle in tools/docker/ — verify the JDBC URL format. The matrix compose.yml shows the Oracle service definition; the per-engine Oracle datasource config is what needs the fix. The Adobe + BoxLang Oracle setups work, so the diff between their configs is the lead.
2. DBMS_LOCK package — needs to be granted to wheelstestdb in the Oracle container's initialization script. Probably an add-grant in tools/docker/oracle/init.sql or equivalent.
3. Constraint cleanup — same shape as the FK collision issue (separate Tier-3 ticket). Migrator test setup needs to drop constraints before re-creating; Oracle's ORA-02264 is the same class of bug as MySQL's Duplicate foreign key.

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 — DB environment, not framework).
• This has been broken on every scheduled matrix run for at least a month (verified back to the 2026-04-12 run).
• Recommend the three sub-clusters be addressed in separate PRs since they're independent: the JDBC URL fix is minutes; the DBMS_LOCK grant requires container rebuild; the constraint cleanup is shared with the MySQL/H2/SQL Server FK issue.
• CLAUDE.md's SOFT_FAIL_DBS guidance is relevant here: while these are being fixed, adding oracle to SOFT_FAIL_DBS in .github/workflows/compat-matrix.yml (currently empty) would let Oracle failures log as warnings rather than hard-fail each engine job.

Related: #2649

[wheels-bot]

Wheels Bot — Triage

Classified as bug.

Suspected layer

migrator — but with a major caveat: the bulk of the work in this issue is CI infrastructure (Docker container init, JDBC URL config, .github/workflows/compat-matrix.yml), which isn't one of the documented framework layers. Only sub-cluster #3 (constraint cleanup in migratorSpec / bulkOperationsSpec) touches vendor/wheels/migrator/ framework code; the other two sub-clusters live in tools/docker/oracle/ and per-engine datasource config under tools/docker/.

Fix sketch

Three independent threads, as the reporter notes:

1. Lucee 6 Oracle JDBC URL — compare the Lucee 6 CFConfig.json Oracle datasource block under tools/docker/ against the working Adobe/BoxLang configs; correct the JDBC URL format. Quick, isolated change.
2. DBMS_LOCK grant — add GRANT EXECUTE ON DBMS_LOCK TO wheelstestdb; (or equivalent) to the Oracle container's init script under tools/docker/oracle/. Requires container rebuild to verify.
3. Constraint cleanup (ORA-02264 / ORA-01003) — same class of bug as the MySQL Duplicate foreign key constraint name issue tracked separately. Test setup needs to drop named constraints before re-creating; likely touches migrator test-helper code or the migrator's drop_table cleanup path. Coordinating fix with the MySQL FK issue is sensible.

Short-term mitigation noted by the reporter: add oracle to SOFT_FAIL_DBS in .github/workflows/compat-matrix.yml (lines 389, 519) so Oracle failures log as warnings rather than hard-fail each engine job while these are fixed.

Confidence: `low`

Three independent sub-clusters in three different files/subsystems — propose-fix cannot sensibly produce one coherent PR. The reporter themselves recommends splitting into three PRs. Auto-downgrade rules also apply (migration files involved, cross-engine concerns), but the primary reason for `low` is the multi-cluster scope: this needs a human to sequence the three threads and decide whether to ship the soft-fail mitigation first.

<!-- wheels-bot:triage:2663 -->
<!-- wheels-bot:triage-class:bug -->