Skip to content

IBX-11939: [Doctrine Migrations] Added Doctrine Migrations-based schema and data install path (inlined SQL) - #785

Open
Steveb-p wants to merge 24 commits into
4.6from
doctrine-migrations-installer-inline-sql
Open

IBX-11939: [Doctrine Migrations] Added Doctrine Migrations-based schema and data install path (inlined SQL)#785
Steveb-p wants to merge 24 commits into
4.6from
doctrine-migrations-installer-inline-sql

Conversation

@Steveb-p

@Steveb-p Steveb-p commented Jul 17, 2026

Copy link
Copy Markdown
Contributor
Question Answer
JIRA issue IBX-11939
Type feature
Target Ibexa version v4.6
BC breaks no

Related PRs:

Package 4.6 5.0 6.0
ibexa/activity-log #166 #167 #168
ibexa/cart #172 #173 🔄 #174
ibexa/collaboration #115 #116
ibexa/connector-ai #198 #199
ibexa/connector-payum ⚠️ #39 #40 #41
ibexa/core #785this PR 🩹 #787 🔄 🩹 #788 🩹
ibexa/corporate-account #369 #370 🔄 #371
ibexa/discounts #341 🩹 #342 🩹
ibexa/discounts-codes #50 #51
ibexa/doctrine-migrations #1 #2 #3
ibexa/doctrine-schema #41 #42 #43
ibexa/fieldtype-page #208 🩹 #209 🔄 🩹 #210 🩹
ibexa/form-builder #240 🩹 #241 🔄 🩹 #242 🩹
ibexa/measurement #133 #134 🔄 #135
ibexa/messenger #22 #23
ibexa/migrations #438 #439 #440
ibexa/oauth2-server 🔄 #48 #49 #50
ibexa/order-management #188 #189 🔄 #190
ibexa/payment #204 #205 🔄 #206
ibexa/product-catalog #1543 #1544 🔄 #1545
ibexa/product-catalog-date-time-attribute #53 #54
ibexa/product-catalog-symbol-attribute #17 #18
ibexa/scheduler #168 #169 🔄 #170
ibexa/segmentation #186 #187 🔄 #188
ibexa/share #190 #191
ibexa/shipping #154 #155 🔄 #156
ibexa/shopping-list #66 #67
ibexa/site-context #121
ibexa/site-factory #172 #173 🔄 🩹 #174 🩹
ibexa/taxonomy ⚠️ #431 #432 #433
ibexa/user #128 #129 🔄 #130
ibexa/workflow #191 🩹 #192 🔄 🩹 #193 🩹

🔄 = this branch adds an upgrade migration (renames/FK-retargets existing schema), not just a fresh baseline.
🩹 = this branch also received a backported delta migration, converted from a legacy ibexa/installer upgrade/db/*.sql script (4.6.0 or later) that the original baseline-only migration didn't cover. See IBX-11939 upgrade-scripts backport report or that package's own PR description for details.
⚠️ = fixes a different, related problem (missing SchemaBuilderEvent support for a plain Doctrine ORM entity/table) — not a Doctrine Migrations baseline/upgrade migration like the rest of this table. See that package's own PR description for details.

Alternative to #784, using the same IbexaMigrationInterface + IbexaMigrationTag + TaggedMigrationsRunner/IbexaOnlyDependencyFactory mechanism, but with InstallSchemaMigration/ImportDataMigration implemented as plain Doctrine\Migrations\AbstractMigration subclasses with all SQL inlined directly in PHP (via addSql()), instead of Ibexa's AbstractVersion + YAML manifest + one-SQL-statement-per-file mechanism. This drops the 509 generated YAML/SQL resource files in favor of two self-contained PHP migration classes.

Adds an ibexa/doctrine-migrations-based alternative to the event-driven SchemaBuilderEvent mechanism CoreInstaller currently uses to install the core schema and bootstrap data, controlled by a new ibexa.installer.schema_builder_event.enabled setting (defaults to true, preserving current behavior).

  • Enabled (default): CoreInstaller keeps dispatching SchemaBuilderEvent and importing cleandata.sql directly - unchanged.
  • Disabled: schema creation and data import are each modeled as a Doctrine migration (InstallSchemaMigration, ImportDataMigration), tagged with IbexaMigrationTag for discovery, and executed via TaggedMigrationsRunner through IbexaOnlyDependencyFactory - an independent Doctrine Migrations DependencyFactory scoped to only Ibexa-tagged migrations, so a project's own migrations are never accidentally picked up. Each migration's execution is recorded in the standard Doctrine Migrations versioning table, making repeated installs idempotent.

InstallSchemaMigration's SQL (mysql/postgresql/sqlite) is generated from the existing schema.yaml; ImportDataMigration's SQL is generated from the existing per-DBMS cleandata.sql files, with the mysql statements also reused verbatim for sqlite (which previously had no cleandata support at all). Long or naturally multi-line statements (CREATE TABLE, multi-row INSERT, long ALTER TABLE ... ADD CONSTRAINT) use PHP NOWDOC for readability; verified byte-for-byte equivalent (aside from whitespace) to the SQL produced by #784.

Requires ibexa/doctrine-migrations (currently tracked via its feat/doctrine-migrations branch, added as a VCS repository).

Checklist:

  • Provided PR description.
  • Tested the solution manually.
  • Provided automated test coverage.
  • Checked that target branch is set correctly (main for features, the oldest supported for bugs).
  • Ran PHP CS Fixer for new PHP code (use $ composer fix-cs).
  • Asked for a review (ping @ibexa/engineering).

@Steveb-p

Copy link
Copy Markdown
Contributor Author

Companion PR in ibexa/installer, adjusting AbstractProductInstaller for CoreInstaller's new constructor arguments: ibexa/installer#213

@Steveb-p
Steveb-p requested a review from a team July 20, 2026 07:34
@ibexa-workflow-automation-1
ibexa-workflow-automation-1 Bot requested review from ViniTou, alongosz, barw4, bnowak, ciastektk, konradoboza, mikadamczyk, tbialcz and wiewiurdp and removed request for a team July 20, 2026 07:34
Comment thread src/bundle/RepositoryInstaller/Command/InstallPlatformCommand.php
Comment thread src/bundle/RepositoryInstaller/Installer/CoreInstaller.php Outdated
Comment thread src/bundle/RepositoryInstaller/Installer/CoreInstaller.php Outdated
Comment thread src/bundle/RepositoryInstaller/Installer/CoreInstaller.php Outdated
Comment thread src/bundle/RepositoryInstaller/Migration/TaggedMigrationsRunner.php Outdated
@Steveb-p
Steveb-p force-pushed the doctrine-migrations-installer-inline-sql branch from 7114546 to 4e9c810 Compare July 24, 2026 11:12
Steveb-p added a commit that referenced this pull request Jul 24, 2026
Replaces the AbstractVersion + YAML manifest + one-SQL-statement-per-file
mechanism with InstallSchemaMigration/ImportDataMigration implemented as
plain Doctrine\Migrations\AbstractMigration subclasses with all SQL inlined
via addSql(), still implementing IbexaMigrationInterface and tagged with
IbexaMigrationTag so TaggedMigrationsRunner/IbexaOnlyDependencyFactory work
unchanged. Long/multiline statements use PHP NOWDOC for readability, with
CREATE TABLE bodies pretty-printed one column per line.

Same conversion as #785 for the 4.6 line, adapted for this
branch's ibexa_*-renamed schema and dbal 3.x platform class names
(AbstractMySQLPlatform/PostgreSQLPlatform instead of MySqlPlatform/
PostgreSqlPlatform).
Steveb-p added a commit that referenced this pull request Jul 24, 2026
Replaces the AbstractVersion + YAML manifest + one-SQL-statement-per-file
mechanism with InstallSchemaMigration/ImportDataMigration implemented as
plain Doctrine\Migrations\AbstractMigration subclasses with all SQL inlined
via addSql(), still implementing IbexaMigrationInterface and tagged with
IbexaMigrationTag so TaggedMigrationsRunner/IbexaOnlyDependencyFactory work
unchanged. Long/multiline statements use PHP NOWDOC for readability, with
CREATE TABLE bodies pretty-printed one column per line.

Same conversion as #785 for the 4.6 line, adapted for this
branch's ibexa_*-renamed schema and dbal 3.x platform class names
(AbstractMySQLPlatform/PostgreSQLPlatform instead of MySqlPlatform/
PostgreSqlPlatform).
Steveb-p and others added 15 commits July 28, 2026 16:17
Symfony\Component\DependencyInjection\ServiceLocator is not a generic
class, so PHPStan rejects the ServiceLocator<Installer> PHPDoc type used
for InstallPlatformCommand::$installers. Removed the invalid generic
parameter; the type-hint itself (plain ServiceLocator) is unaffected.
Converts InstallSchemaMigration and ImportDataMigration from
Ibexa\Contracts\DoctrineMigrations\Migrations\AbstractVersion (which loads
its SQL from a YAML manifest + one-statement-per-file pairs at runtime) to
Doctrine\Migrations\AbstractMigration directly, with every statement
inlined as an addSql() call in up(), branching on $this->platform
(MySqlPlatform/PostgreSqlPlatform/SqlitePlatform).

Both still implement IbexaMigrationInterface and are tagged with
IbexaMigrationTag::TAG in services.yml, so they remain discoverable by
TaggedMigrationsRunner/ServiceMigrationsRepository/IbexaOnlyDependencyFactory
exactly as before - no other part of the installer changes.

Removes the entire Resources/migrations/ YAML+SQL-file tree (509 files),
generated from the same schema.yaml/cleandata.sql sources as before.
CREATE TABLE statements are now pretty-printed one column per line, and
other long or naturally multi-line statements (multi-row INSERT, long
ALTER TABLE ... ADD CONSTRAINT) use NOWDOC instead of a single escaped
string literal, for readability. SQL content is unchanged.
Extends AbstractSqlMigration (added in ibexa/doctrine-migrations) instead of the plain
Doctrine AbstractMigration, replacing `$this->platform instanceof ...` checks with
isMySQL()/isPostgreSQL()/isSqlite(), and moving each platform Statement block out of the
PHP file into its own sql/*.sql file loaded via addSqlFile().

Mechanical, content-preserving change: every migration was run before and after against
all three platforms and the resulting SQL statement lists are byte-for-byte identical.
Call abortIfUnsupportedPlatform() as the first statement of up(), so
installs on a database this migration doesn't build SQL for fail loudly
instead of silently queuing zero statements.
Each statement now ends with `;`, matching ibexa:doctrine:schema:dump-sql's
own convention, so the files are directly executable via mysql/psql/sqlite3
CLI clients. addSqlFile() still splits on the delimiter and passes one
statement per addSql() call, unaffected by the trailing terminator.
…tions

Guards ibexa/core's InstallSchemaMigration/ImportDataMigration/
RenameSchemaTo5_0Migration and ibexa/fieldtype-page's
RenameSchemaTo5_0Migration so they skip when an existing install already
built its schema the old way. ImportDataMigration checks for the absence
of its pre-rename input table, since migrations run in a strict,
deterministic order -- if it's missing, this system never went through
this migration, meaning it built its schema (and bootstrap data) via
schema.yaml directly.

The two data-value fixes hidden inside the rename migrations (core's
ez_lock/ezstring identifier fixes, fieldtype-page's block-ID HTML
reference fix) are split into their own always-run migrations
(FixLegacyIdentifiersMigration, FixBlockIdReferencesMigration), since a
schema-shape guard on the rename would otherwise also skip these on a
system whose schema is already renamed but whose row data still holds
the old values -- schema.yaml only defines structure, not content. Both
run unconditionally right after their corresponding rename (same target
version, later creation date), so the tables they touch are always at
their final name by then, and each statement is a no-op via its own
WHERE clause once already applied.
Doctrine Migrations only calls MetadataStorage::complete() (the write to
doctrine_migration_versions) when a migration's up() returns normally --
never when it throws SkipMigration. So every migration using skipIf() was
being silently re-evaluated on every future doctrine:migrations:migrate
run instead of being permanently recorded as applied, even though its
guard condition (the schema already being in place) never changes back.

Replaces every `$this->skipIf($condition, $message);` with
`if ($condition) { return; }`: up() now returns normally with zero queued
SQL when the guard fires, so the migration is correctly recorded as
executed (with a "did not result in any SQL statements" warning logged,
which is expected and harmless) and never re-evaluated again.

Verified end-to-end: fresh ibexa:install (schema_builder_event enabled)
followed by doctrine:migrations:migrate now records all 44 tagged
migrations in doctrine_migration_versions in one pass, and a second
migrate run does zero work at all ("Already at the latest version").
…name

Live end-to-end testing (fresh legacy install -> doctrine:migrations:migrate
on 4.6) surfaced that these baseline guards checked the FINAL, post-5.0-
rename table name (e.g. "ibexa_content") -- correct on 5.0/6.0, but wrong
on 4.6, which has no rename at all: there, the table this migration
creates already IS the branch's permanent, current name (e.g.
"ezcontentobject"), so the guard never fired and the baseline collided
with an already-installed legacy 4.6 schema. Checks the branch's own
table name instead, exactly like every other baseline guard that has no
later rename to worry about.

For core's ImportDataMigration specifically, this required a proper data
check rather than a schema-shape check: "ezcontentobject" is 4.6's real,
permanent name, so it always exists once the baseline has run there,
regardless of whether this migration's own bootstrap INSERTs ran yet.
Checks for the actual seed row (root content, id = 1) instead, combined
with the table's outright absence (the 5.0/6.0 legacy signal) via OR.
…emaBuilderEvent

ibexa:doctrine:migrations:diff (and any other Doctrine Migrations command that
needs a target schema) previously failed with 'The schema provider is not
available.', since IbexaOnlyDependencyFactory never had an entity manager or
SchemaProvider configured.

SchemaBuilderEventSchemaProvider bridges Doctrine Migrations' SchemaProvider
to Ibexa's own legacy, event-driven schema builder (SchemaBuilderInterface,
backed by SchemaBuilderEvent and every installed package's
BuildSchemaSubscriber) -- the same mechanism CoreInstaller::importSchema()
already uses for the legacy install path. A new compiler pass wires it onto
IbexaOnlyDependencyFactory via setService(), a no-op if ibexa/doctrine-schema
or ibexa/doctrine-migrations isn't installed.
…nt importData() call)

- CoreInstaller::getQueriesFromSchemaBuilderEvent() now builds its Query list via an
  arrow function instead of a static closure, per review suggestion.
- CoreInstaller::importData() no longer re-invokes TaggedMigrationsRunner::run() in the
  migrations-runner branch; importSchema() already runs every tagged migration
  (including ImportDataMigration) to the latest version, so the second call was a
  pure no-op with no real caller relying on it standalone.
- getDropSqlStatementsForExistingSchema()'s docblock tightened to list<string>, which
  it always genuinely returns.
- TaggedMigrationsRunner's constructor documents why $dependencyFactory is nullable
  (optional service, wired via "@?..." in services.yml; run() throws a clear exception
  if called while null).
ibexa/doctrine-migrations is now published on Packagist (which mirrors all
of its branches, not just tags), so the explicit VCS repository pointing
composer directly at GitHub is no longer needed to resolve the dev-branch
require constraint.
… mandatory

Instead of TaggedMigrationsRunner tolerating a null DependencyFactory at
runtime, RemoveTaggedMigrationsRunnerPass now removes its service
definition entirely when "ibexa/doctrine-migrations" isn't
installed/enabled, and CoreInstaller depends on it as an optional
(nullable) service instead, throwing a clear configuration exception if
"ibexa.installer.schema_builder_event.enabled" is disabled without it.
…erEvent()

Passing an empty array as array_map()'s third argument re-indexes its
result, so PHPStan can type it as list<Query> instead of Query[].
@Steveb-p
Steveb-p force-pushed the doctrine-migrations-installer-inline-sql branch from 799cb50 to cad60ff Compare July 28, 2026 14:50
Steveb-p added 7 commits July 29, 2026 00:25
Converts installer's upgrade/db/ibexa-4.5.1-to-4.5.2.sql (indexes on
ezcontentobject_link, ezcontentclass_attribute, ezurl_object_link,
ezcontentobject_attribute) and upgrade/db/ibexa-4.6.20-to-4.6.21.sql
(ezurlalias_ml "link" index) into guarded Doctrine migrations, for
installs whose schema already predates these indexes.
$schema->getTable() throws TableDoesNotExist rather than returning null,
so a guard that calls it directly (without checking hasTable() first)
crashes -- rather than skipping -- on any install that never had the
table under this name at all (e.g. a database built directly at the
current schema, which never passed through this pre-rename/pre-delta
state). Caught via live testing against a real database. Guard now
checks hasTable() first.
…into baseline

Same rationale as ibexa/product-catalog: doctrine migrations only exist
starting at 4.6, so InstallSchemaMigration's baseline already includes
these indexes (originally shipped via installer's
upgrade/db/ibexa-4.5.0-to-4.5.1.sql). A standalone migration for pre-4.6
content can never have real work to do on a genuine 4.6+ install.
The feat/doctrine-migrations branch it previously required has been
merged into and deleted from the primary 4.6 branch upstream, so the
old constraint no longer resolves.
$migration->up(new Schema()) gave every migration's hasTable()/
getTable() guard checks an always-empty schema to inspect, regardless
of what earlier migrations in the same run (or a prior run) actually
created against the real database -- guards written the correct way
(checking hasTable() before trusting a table exists) would incorrectly
conclude the table is missing and either skip work that still needs
doing, or crash outright on getTable() for a table that does exist.
Introspect the live database instead, the same way the standalone
'doctrine:migrations:migrate' command already does via
DBALSchemaDiffProvider::createFromSchema().
InstallSchemaMigration, AddUrlAliasMlLinkIndexMigration, and ImportDataMigration
only depend on the persistence connection -- they don't need anything from
IbexaRepositoryInstallerBundle (CoreInstaller, BuildSchemaSubscriber), which is
scoped to installer-time concerns and additionally requires DoctrineSchemaBundle
to boot at all. Declaring them in IbexaCoreBundle instead matches how every
other package in this effort declares its own InstallSchemaMigration in its own
bundle, and lets any consumer discover core's schema migrations via
ibexa:doctrine:migrations:migrate without needing the installer bundle
registered -- useful for integration test kernels that don't otherwise need
CoreInstaller.
… SQL

install-schema-sqlite.sql and import-data-sqlite.sql were generated from
the MySQL source without adapting to SQLite's stricter SQL semantics:

- import-data-sqlite.sql used MySQL-style backslash escapes (\" and \n)
  inside string literals. SQLite (unlike MySQL) does not interpret
  backslash as an escape character at all, so these were stored as
  literal two-byte sequences, corrupting PHP-serialized content-type
  names/descriptions (unserialize() failures) and embedded XML field
  data (DOMDocument parse failures, e.g. the admin user's own ezimage
  field).

- install-schema-sqlite.sql collapsed the composite primary keys on
  ezcontentclass, ezcontentclass_attribute and ezcontentobject_attribute
  (id, version) down to a single-column `id INTEGER PRIMARY KEY
  AUTOINCREMENT`, dropping the `version` component entirely -- unlike
  the mysql/postgresql variants, which both correctly declare
  `PRIMARY KEY(id, version)`. This made SQLite reject the second row
  (e.g. publishing a content-type draft, which legitimately reuses the
  same id with a different version/status) as a duplicate primary key.

Both only affect the SQLite platform; MySQL and PostgreSQL were
already correct.
@Steveb-p

Copy link
Copy Markdown
Contributor Author

Opened ibexa/doctrine-migrations#4, fixing the RegisterMigrationsPass container-compilation failure (doctrine.migrations.logger non-existent service) causing this PR's Behat CI failure. Once that lands and the ibexa/doctrine-migrations branch this PR pins is updated, the Behat job should pass.

…nstanceOf ignores

Unblocks this PR's CI. Pre-existing on 4.6 itself (unrelated to this PR's
changes) - see #794, which fixes it at the source; this commit
will become a no-op once that's merged and this branch is rebased on top.
@Steveb-p

Copy link
Copy Markdown
Contributor Author

Also pushed a temporary fix for the PHPStan baseline drift (LocationServiceTest.php assertInstanceOf counts) blocking this PR's CI — it's pre-existing on 4.6 itself, unrelated to this PR. Filed the real fix separately as #794; once that merges and this branch rebases, this extra commit becomes a no-op.

@sonarqubecloud

Copy link
Copy Markdown

Quality Gate Failed Quality Gate failed

Failed conditions
17.1% Duplication on New Code (required ≤ 3%)
C Reliability Rating on New Code (required ≥ A)
E Security Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants