IBX-12603: Resolved Ibexa DB platform when generating test database schema by alongosz · Pull Request #52 · ibexa/test-core

alongosz · 2026-09-15T10:29:30Z

🎫 Issue	IBX-12603

Related PRs:

IBX-12043: Upgraded to Doctrine DBAL 4 core#800
IBX-12530: Rebuilt DatabaseSchemaHook on SchemaBuilderInterface instead of a raw-file reimplementation #43
IBX-12530: Added SchemaApplier for executing a built Schema against a connection doctrine-schema#49

Description:

DatabaseSchemaHook generates DDL with $connection->getDatabasePlatform(), which since the DBAL 4 upgrade returns vanilla Doctrine\DBAL\Platforms\SQLitePlatform instead of our SqliteDbPlatform. Vanilla SQLitePlatform drops the table-level PRIMARY KEY (...) clause as soon as a table has an autoincrement column, so composite keys silently degrade to a single column:

SQLitePlatform:   CREATE TABLE ibexa_content_field (id INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL, version ...)
SqliteDbPlatform: CREATE TABLE ibexa_content_field (id INTEGER NOT NULL, version ..., PRIMARY KEY (id, version))

ibexa_content_field is keyed on (id, version), one row per content version, so the fixture repeats id on purpose and the import dies in bootstrap with UNIQUE constraint failed: ibexa_content_field.id. ibexa_content_type and ibexa_content_type_field_definition lose their (id, status) key the same way. This is what makes ibexa/core 6.0 red: run 34885080403.

SQLite only, 6.0 only. It took two changes to get here:

IBX-12043: Upgraded to Doctrine DBAL 4 core#800 removed doctrine-bundle's platform_service, gone in DBAL 4. It compensated in CoreInstaller and LegacySchemaImporter, but this hook was missed. 4.6 and 5.0 are on DBAL 3 and still wire platform_service, so the connection returns the right platform there.
IBX-12530: Rebuilt DatabaseSchemaHook on SchemaBuilderInterface instead of a raw-file reimplementation #43 rebuilt this hook on SchemaBuilderInterface. Before that it delegated to LegacySchemaImporter, which resolves the platform via DbPlatformFactory.

Resolved the platform via DbPlatformFactoryInterface instead of taking it off the connection, the same way CoreInstaller and LegacySchemaImporter already do
[Tests] Added DatabaseSchemaHookTest - builds a schema shaped like ibexa_content_field and asserts the emitted DDL keeps PRIMARY KEY (id, version). Confirmed it fails against the previous code.
[CS] Overrode php_unit_test_case_static_method_calls for createStub only - it is not static in PHPUnit 9, so $this-> is the accurate form and PHPStorm flags self::. Bumped ibexa/code-style to ~2.3.0 while at it.
🟢 composer integration on ibexa/core 6.0 with this branch symlinked in: legacy suite 11417 tests green, phpunit-integration.xml 79 tests green (was dying in bootstrap)
🟢 composer check-cs, composer test, composer phpstan green here

The test is deliberately a tripwire for the note below rather than broad coverage - it fails if this call site ever goes back to the connection's platform. Real integration coverage of SqliteDbPlatform DDL generation belongs in ibexa/doctrine-schema, not here.

Note

ibexa/doctrine-schema#49 moves this call site onto a new SchemaApplier, which carries the same $connection->getDatabasePlatform(), plus Connection::getSchemaManager() and getDropTableSQL(Table) - both gone in DBAL 4. That chain is based on 4.6, so all three need handling when it merges up into 6.0, otherwise this comes back.

For QA:

No QA required. CI covers it:

🟢 [TMP] IBX-12603: Pointed CI at ibexa/test-core#52 core#830
🟢 ibexa/taxonomy#452

Documentation:

No documentation required.

🤖 Generated with Claude Code

DatabaseSchemaHook generated DDL with $connection->getDatabasePlatform(), which returns vanilla Doctrine SQLitePlatform. That platform drops the table-level PRIMARY KEY clause whenever a table has an autoincrement column, so composite keys degraded to a single column and fixture import failed on ibexa_content_field(id, version). DBAL 4 removed doctrine-bundle's platform_service, so the connection no longer carries Ibexa's platform subclasses. Resolve them explicitly via DbPlatformFactoryInterface, the same way CoreInstaller and LegacySchemaImporter already do. Affects 6.0 only. 4.6 and 5.0 are on DBAL 3 and still wire platform_service. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

6.0 integration test bootstrap fails on SQLite because DatabaseSchemaHook generates DDL with the vanilla Doctrine platform, losing composite primary keys. The fix lives in ibexa/test-core, nothing in this repository changes. dependencies.json makes CI install that branch, so the fix can be proven here before it merges. Generated with ci dependencies:link, must be dropped once ibexa/test-core#52 is merged. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Builds a schema with a composite primary key over an autoincrement column, the shape ibexa_content_field has, and asserts the emitted DDL keeps PRIMARY KEY (id, version). Fails against the previous code, which took the platform off the connection. This is the guard for the SchemaApplier work on 4.6: if that call site goes back to $connection->getDatabasePlatform() when the chain merges up, this test fails instead of the bug returning silently. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

php_unit_test_case_static_method_calls rewrites $this->createStub() to self::, which contradicts PHPStorm's EA inspection. createStub is a non-static method in PHPUnit 9, so $this-> is the accurate form here. The fixer matches on method name across PHPUnit versions and assumes PHPUnit 10+, where createStub did become static, hence the disagreement. Overriding the call type for that one method keeps the rest of the suite on self::. Revisit when moving to PHPUnit 10+, where the override stops being correct. Bumped ibexa/code-style to ~2.3.0 while at it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

alongosz · 2026-09-15T12:35:58Z

            'inline_constructor_arguments' => false,
        ],
-        'php_unit_test_case_static_method_calls' => ['call_type' => 'self'],
+        'php_unit_test_case_static_method_calls' => ['call_type' => 'self', 'methods' => ['createStub' => 'this']],


Note
Well... That fixer is of... questionable quality. It hard-codes list of method it thinks should be static rather than inferring from the installed source (maybe can't?). PHPUnit methods' signatures change across different PHPUnit version. Right now createStub is not static (PHPUnit 9).

maybe that should be delegated to phpstan? tbh, fixer saying that something is static or not seems kinda too much.

maybe that should be delegated to phpstan? tbh, fixer saying that something is static or not seems kinda too much.

Yeah, not sure why it was enabled here. It belongs to the @PhpCsFixer:risky rule set, not enabled by default.
I can drop it completely.

POV ping @Steveb-p

alongosz · 2026-09-15T12:36:30Z

    "require-dev": {
        "phpunit/phpunit": "^9",
-        "ibexa/code-style": "^2.0",
+        "ibexa/code-style": "~2.3.0",


Bumped while at it, causes no CS drifts.

ViniTou · 2026-09-15T12:42:22Z

            'inline_constructor_arguments' => false,
        ],
-        'php_unit_test_case_static_method_calls' => ['call_type' => 'self'],
+        'php_unit_test_case_static_method_calls' => ['call_type' => 'self', 'methods' => ['createStub' => 'this']],


maybe that should be delegated to phpstan? tbh, fixer saying that something is static or not seems kinda too much.

Steveb-p

Good enough to unblock.

alongosz mentioned this pull request Sep 15, 2026

[TMP] IBX-12603: Pointed CI at ibexa/test-core#52 ibexa/core#830

Closed

2 tasks

alongosz and others added 2 commits September 15, 2026 14:15

alongosz commented Sep 15, 2026

View reviewed changes

alongosz requested review from Steveb-p, ViniTou, barw4, bnowak, ciastektk, konradoboza, mikadamczyk, tbialcz and wiewiurdp September 15, 2026 12:37

konradoboza approved these changes Sep 15, 2026

View reviewed changes

ViniTou approved these changes Sep 15, 2026

View reviewed changes

Steveb-p approved these changes Sep 15, 2026

View reviewed changes

bnowak approved these changes Sep 15, 2026

View reviewed changes

alongosz merged commit d78ca95 into 6.0 Sep 15, 2026
18 checks passed

alongosz deleted the ibx-12603-sqlite-composite-pk branch September 15, 2026 12:59

Navigation Menu

IBX-12603: Resolved Ibexa DB platform when generating test database schema - #52

IBX-12603: Resolved Ibexa DB platform when generating test database schema#52
alongosz merged 3 commits into
6.0from
ibx-12603-sqlite-composite-pk

alongosz commented Sep 15, 2026 •

edited

Loading

Uh oh!

alongosz Sep 15, 2026

Uh oh!

ViniTou Sep 15, 2026

Uh oh!

alongosz Sep 15, 2026

Uh oh!

alongosz Sep 15, 2026

Uh oh!

ViniTou Sep 15, 2026

Uh oh!

Steveb-p left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

alongosz commented Sep 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Related PRs:

Description:

For QA:

Documentation:

Uh oh!

alongosz Sep 15, 2026

Choose a reason for hiding this comment

Uh oh!

ViniTou Sep 15, 2026

Choose a reason for hiding this comment

Uh oh!

alongosz Sep 15, 2026

Choose a reason for hiding this comment

Uh oh!

alongosz Sep 15, 2026

Choose a reason for hiding this comment

Uh oh!

ViniTou Sep 15, 2026

Choose a reason for hiding this comment

Uh oh!

Steveb-p left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

alongosz commented Sep 15, 2026 •

edited

Loading