fix(sqlite-kit): preserve NOT NULL on non-integer primary key columns

[webdevsamran]

Summary

Fixes #6165.

On the v1 (beta) line, drizzle-kit generate drops NOT NULL from every primary-key column when serializing a Drizzle schema into the interim/snapshot representation:

• drizzle-kit/src/dialects/sqlite/drizzle.ts: notNull: column.notNull && !primaryKey
• drizzle-kit/src/cli/commands/up-sqlite.ts (v6→v7 upgrader): notNull: column.notNull && !column.primaryKey

Since SQLite only enforces non-NULL values on INTEGER PRIMARY KEY columns (rowid alias), NOT NULL is load-bearing for every other PK shape — text('id').primaryKey() produced:

CREATE TABLE `users` (
	`id` text PRIMARY KEY
);

which silently allows NULL ids — a regression from 0.31 behavior and a data-integrity hazard for anyone who doesn't hand-patch their migrations.

The DDL convertor (drizzle-kit/src/dialects/sqlite/convertor.ts) already implements the correct SQLite-specific rule: it omits NOT NULL only for single-column integer PKs (omitNotNull = isColumnPk && column.type.toLowerCase().startsWith('int')). The serializer stripping the flag upstream of it made that logic unreachable for PK columns. This change makes the serializer faithful (notNull: column.notNull) and lets the convertor keep deciding what to emit, which:

• restores NOT NULL for TEXT/BLOB/REAL/INT PKs and any explicitly .notNull() PK column,
• keeps omitting redundant NOT NULL for single-column integer().primaryKey() (the behavior requested in [FEATURE]: SQLite .primaryKey().notNull() #2611),
• keeps upgraded (v6→v7) snapshots consistent with freshly generated ones.

Reproduction (before)

With drizzle-orm@1.0.0-rc.4 + drizzle-kit@1.0.0-rc.4:

// schema.ts
import { sqliteTable, text } from 'drizzle-orm/sqlite-core';
export const users = sqliteTable('users', { id: text('id').primaryKey() });

npx drizzle-kit generate → `id` text PRIMARY KEY (no NOT NULL; verified: SQLite accepts INSERT INTO users VALUES (NULL)).

Testing

New regression tests in drizzle-kit/tests/sqlite/sqlite-pk-notnull.test.ts:

• text PK keeps notNull: true in the serialized schema
• generated SQL emits `id` text PRIMARY KEY NOT NULL for text PKs
• generated SQL still emits `id` integer PRIMARY KEY without NOT NULL for single-column integer PKs

✓ tests/sqlite/sqlite-pk-notnull.test.ts (3 tests) 5ms

 Test Files  1 passed (1)
      Tests  3 passed (3)

Verified red→green: with the fix stashed, 2 of the 3 new tests fail; with the fix applied all pass. Full tests/sqlite/ suite comparison on this machine (Windows checkout lacks a better-sqlite3 prebuilt for Node 24, so DB-backed suites fail identically before and after the change): clean beta → 23 failed / 21 passed; with this change → 21 failed / 23 passed (+2 = the new regression tests; no previously passing test regressed).

oxlint clean on all changed files.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0c9191c0d1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Emit a migration for existing nullable primary keys

For projects whose existing v7 snapshot or database was produced by the buggy release, this changes the desired text-PK column from notNull: false to true, but ddlDiff unconditionally deletes every single-column primary-key notNull alteration in drizzle-kit/src/dialects/sqlite/diff.ts:259-260. Both generate and push therefore emit no table recreation, and generate records the new snapshot as non-null despite leaving the actual column nullable, so affected databases are never repaired. Restrict that suppression to actual INTEGER PRIMARY KEY rowid aliases so non-integer PKs receive a corrective migration.

Useful? React with 👍 / 👎.