[bpamiri]

Fixes #4373.

delete(callbacks = false) covers dependent records

hasMany() / hasOne() with dependent = "delete" or "remove" instantiate the child records and delete or update them. delete(callbacks = false) on the parent didn't pass callbacks on, so the children's callbacks still ran. $deleteDependents() now takes callbacks and passes it to every dependent call, and $delete() passes its own.

The records those calls load were also found without callbacks. deleteAll(instantiate = true), deleteOne() and updateOne() called findAll() / findOne() without it, so afterFind / afterInitialization still ran with callbacks = false. They now pass it on, as updateAll(instantiate = true) already did.

For hasMany, dependent = "deleteAll" / "removeAll" work in bulk and run no callbacks. For hasOne, they go through deleteOne() / updateOne(), which load the record, and got no callbacks argument at all. They now get one too.

An invalid dependent value fails when declared

An unsupported value, such as dependent = "destroy", was accepted by hasMany() / hasOne() and only threw Wheels.InvalidArgument at the first delete. It now throws the same error, with the same message, when the association is declared. Valid values are unchanged: delete, deleteAll, remove, removeAll and false.

An app that declared an invalid value and never deleted through that association used to load; it now fails at model load, in every environment.

Specs

vendor/wheels/tests/specs/model/dependentCallbacksSpec.cfc uses dedicated test models, so no shared class changes. They are DepPost, with beforeDelete / beforeSave logging to the request, and three parent models: dependent = "delete", "remove" and "destroy". DepPost logs beforeDelete, beforeSave, afterFind and afterRollback. The parent fixtures cover hasMany (delete, remove, destroy) and hasOne (all four values). Each delete runs with transaction = "rollback".

• callbacks = false with hasMany dependent = "delete" / "remove": no child callback runs (find, delete, save or rollback).
• callbacks = false with hasOne and each of delete, deleteAll, remove, removeAll: no child callback runs.
• callbacks = true: the children's afterFind, write and afterRollback callbacks run, for hasMany and for a hasOne with deleteAll.
• dependent = "destroy" throws Wheels.InvalidArgument when the model loads.
• Every documented value is accepted.

RED on develop: the two callbacks = false cases (3 child callbacks ran), the declaration case and the accepted-values case fail. The callbacks = true control passes.

GREEN, Lucee 7 + SQLite: full suite 6943/6943; complexity gate passes. The hasOne and finder-callback cases were RED before the second commit (afterFind ran).

🤖 Generated with Claude Code

[bpamiri]

REQUEST CHANGES at 4a8ee44:

• MEDIUM — vendor/wheels/model/associations.cfm:295: hasOne uses deleteObject/removeObject, which resolve to deleteOne/updateOne even for dependent="deleteAll"/"removeAll". These branches therefore instantiate a child, but still omit the parent's callbacks flag. Pass it through and cover hasOne write and transaction callbacks for false/true. The PR body's statement that these modes never instantiate needs qualification.
• MEDIUM — vendor/wheels/model/delete.cfm:39, :147, and vendor/wheels/model/update.cfm:206: the internal findAll/findOne calls omit callbacks, so child afterFind/afterInitialization can still run during a dependent operation with callbacks=false. Forward the flag through child loading and add finder-callback regressions alongside the existing write-callback checks.

The new declaration validator preserves case-insensitive operation names and boolean-false values by inspection. The current regression suite covers hasMany beforeDelete/beforeSave in rollback mode; it does not yet prove the single-record or transaction-callback paths above. Final gate remains the revised-head matrix plus PR CI.

[bpamiri]

Source APPROVE at ab06ed7; final gate pending.

Both propagation gaps are resolved: all dependent branches now pass callbacks, including the single-record paths behind hasOne bulk-named modes, and the three internal child-loading queries pass it into their finders. The new fixtures cover all four hasOne dependent modes with callbacks disabled; positive controls verify finder, delete and rollback callbacks still run when enabled.

Evidence qualification: the suite directly tests afterRollback, not afterCommit; suppression of both follows the shared transaction-callback enqueue guard. Likewise, afterFind is directly logged while initialization suppression follows the same forwarded finder flag. No remaining source blocker. Final approval awaits the revised-head matrix and PR CI.

[github-actions]

Wheels Test Results

28 744 tests   28 164 ✅  9m 31s ⏱️
 2 276 suites     580 💤
     4 files         0 ❌

Results for commit a845fc6.

♻️ This comment has been updated with latest results.

[bpamiri]

rev1-r1, taking the final over from rev1-r2. Delta check of the develop merge with #4397: head a845fc6. The conflict resolution is correct.

I compared this PR's own diff against its merge base, before (ab06ed7, which rev1-r2 source-approved) and after (a845fc6). Every difference is in the expected place:

• $deleteDependents now takes both callbacks (this PR) and parentSoftDeleted (fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397). The body keeps fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397's $dependentKeptBySoftDelete gate and forwards callbacks in all four dependent= branches (delete, remove, deleteAll, removeAll).
• In fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397's $deleteRecordAndDependents(), both call sites pass callbacks = arguments.callbacks: the soft path (parentSoftDeleted = true) and the permanent path.
• fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397's $delete() wrapper (catch-free finally, soft-delete restore) is unchanged.
• The rest of the PR diff (the update.cfm line, the Dep* fixtures, dependentCallbacksSpec) is byte-identical to ab06ed7.

Full matrix 37379675857 (on ab06ed7): 32/32 legs green, dependentCallbacksSpec 6/6 on every leg. The final follows the narrowed runs on this head (37397511929, Lucee 7 × all 7 databases; 37397521096, 4 engines × SQLite) and PR CI.

[bpamiri]

FINAL: APPROVE. Head a845fc6 (ab06ed7 + the develop merge with #4397). rev1-r1, taking over from rev1-r2.

• Full compat matrix 37379675857 (on ab06ed7, the code rev1-r2 source-approved): 32/32 legs green, dependentCallbacksSpec 6/6 on every leg.
• Develop merge / conflict resolution (delta-checked in my earlier comment):
  • $deleteDependents(softDelete, includeSoftDeletes, callbacks, parentSoftDeleted) keeps fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397's soft-delete gate and forwards callbacks in all four dependent= branches.
  • Both fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397 call sites pass callbacks, and fix(model): keep soft deletes reversible and make softDelete=false a real delete #4397's $delete() wrapper is unchanged.
  • The rest of the PR diff is byte-identical to ab06ed7.
• Narrowed 37397511929, Lucee 7 × all 7 databases: 7/7 green. dependentCallbacksSpec 6/6 and SoftDeleteSemanticsSpec 18/18 on every database, so both PRs' delete-path behaviours pass together, including on Oracle and CockroachDB.
• Narrowed 37397521096, the other 4 engines × SQLite: 4/4 green, both specs green on Lucee 6, Adobe 2023, Adobe 2025 and BoxLang.
• PR CI on a845fc6: 18 pass, 0 fail, 0 pending, 3 skipping.