refactor(schema): treat srid, dimensions and auto-increment as type parameters

[abnegate]

Stacked on #18 → #17 → #16 → #15. Closes the entanglement I flagged as out of scope there.

The blocker, and why it dissolved

These three modifiers were dropped by dialects that can't emit them, but I couldn't move them in #18 because the library's own factories called them:

Table::point()        -> $col->srid($srid);
Table::id()           -> $col->autoIncrement();
Trait\Serial          -> $col->autoIncrement();
Table\*::vector()     -> $col->dimensions($dimensions);

Moving the modifier would have broken the factory on exactly the dialects that needed it.

The block turned out to be framing. srid and dimensions aren't modifiers at all — they're intrinsic to the column, the way length is. POINT SRID 3857, VECTOR(384): both are part of the column type, and the factories already take them as arguments. So they belong in the constructor next to length/precision/scale, threaded through newColumn(), with the auto-increment flag beside them.

Once the factories stop going through the public modifiers, the modifiers are free to live only where the value is emitted:

Modifier	Now on	Emits
srid()	MySQL, MariaDB, PostgreSQL	POINT SRID 3857, GEOMETRY(POINT, 3857)
dimensions()	PostgreSQL	VECTOR(384)
autoIncrement()	MySQL, MariaDB, PostgreSQL, SQLite	AUTO_INCREMENT, IDENTITY, AUTOINCREMENT

Callers are unaffected

point($name, $srid), id() and serial() still work on every dialect and still record the values — only the standalone modifiers are scoped. Verified byte-identical DDL against the pre-change output for all five dialects:

MySQL       id()     : CREATE TABLE `t` (`id` BIGINT UNSIGNED AUTO_INCREMENT NOT NULL, PRIMARY KEY (`id`))
PostgreSQL  point()  : CREATE TABLE "t" ("g" GEOMETRY(POINT, 3857) NOT NULL)
SQLite      serial() : CREATE TABLE `t` (`s` INTEGER AUTOINCREMENT NOT NULL)
ClickHouse  id()     : CREATE TABLE `t` (`id` UInt64) ENGINE = MergeTree() ORDER BY (`id`)
PostgreSQL  vector() : CREATE TABLE "t" ("e" VECTOR(384) NOT NULL)

A test pins both halves: the modifier is absent on ClickHouse/MongoDB/SQLite while id()/serial()/point() still set isAutoIncrement and srid there.

One signature change

Table\ClickHouse::vector() and Table\MongoDB::vector() lose their $dimensions parameter. Array(Float64) and the array bsonType are unsized, so it could never be honoured — and two existing tests passed 768 while asserting output that ignored it:

->vector('embedding', 768)
// CREATE TABLE `embeddings` (`embedding` Array(Float64)) ENGINE = MergeTree() ...
//                                        ↑ 768 nowhere

That's the clearest evidence it was never wired to anything. Table\PostgreSQL::vector($name, $dimensions) keeps it, because PostgreSQL emits it.

Left as the one known gap

unsigned() stays on every dialect. It renders nothing on PostgreSQL and SQLite, but through an explicit overridable compileUnsigned() hook returning '' — a deliberate dialect mapping, not an oversight. It's the single remaining entry in the README's "Known gaps" note.

Test plan

• composer test — 5323 tests, 12383 assertions, all pass (was 5321)
• composer check — PHPStan level max, no errors
• composer lint — pass
• DDL output diffed against pre-change for all five dialects across id(), point(), serial(), vector() — identical

Integration tests need Docker and were not run locally; CI covers them.

🤖 Generated with Claude Code

[github-actions]

📊 Coverage

Metric	Covered	Ratio
Lines	91.87%	7454 / 8114
Methods	84.13%	1108 / 1317
Classes	65.45%	144 / 220

Full per-file breakdown in the job summary.

[greptile-apps]

Greptile Summary

The PR moves SRID, vector dimensions, and auto-increment state into column construction while limiting their standalone modifiers to dialects that emit them.

• Threads intrinsic type state through Table::newColumn() and dialect overrides.
• Adds dialect-specific SRID, dimensions, and auto-increment traits.
• Removes unused vector dimensions from ClickHouse and MongoDB APIs and updates documentation and focused tests.

Confidence Score: 4/5

The PR should not merge until the base fluent-builder test is updated or the base Column autoIncrement() contract is restored.

The changed base Column no longer exposes autoIncrement(), while an unchanged test directly invokes that method on a base Column and will fail with an undefined-method error.

Files Needing Attention: src/Query/Schema/Column.php, tests/Query/Schema/FluentBuilderTest.php

Important Files Changed

Filename	Overview
src/Query/Schema/Column.php	Adds intrinsic constructor state but removes a base fluent method that an unchanged test still invokes.
src/Query/Schema/Table.php	Threads SRID and auto-increment state directly through newColumn() for generic factories.
src/Query/Schema/Table/PostgreSQL.php	Preserves PostgreSQL vector dimensions through the expanded column factory contract.
src/Query/Schema/Table/ClickHouse.php	Removes the previously ignored vector-dimensions parameter and forwards the expanded constructor state.
src/Query/Schema/Table/MongoDB.php	Removes the previously ignored vector-dimensions parameter and retains intrinsic factory state.
tests/Query/Schema/FluentBuilderTest.php	Adds dialect-scoping assertions but leaves an existing base-Column autoIncrement() invocation that now fails.

Prompt To Fix All With AI

### Issue 1
src/Query/Schema/Column.php:73-82
**Base auto-increment method removed**

When `FluentBuilderTest::testColumnReturnsItselfForChainingFluentMethods` creates a base `Column`, the test still calls `autoIncrement()`, but this change removes that method from the base class, causing the test suite to terminate with an undefined-method error.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{Reviews (1): Last reviewed commit: "refactor(schema): treat srid, dimensions..." | Re-trigger Greptile}