[PerryLink]

SQLiteDriver.start() at packages/sqlite/src/index.ts:166 parses config.path as a URL:

this.path = fileURLToPath(new URL(this.path, this.ctx.baseUrl))

path is documented as a filesystem path (z.string().role('path'), locale path: Database path.), but on Windows new URL('C:\\data\\koishi.db', base) reads C: as a URL scheme, and fileURLToPath throws TypeError: The URL must be of scheme file. An absolute path — accepted by resolve() before the migration — therefore cannot start the driver on Windows.

This is a regression from the v4 migration. origin/v3 resolved it as a filesystem path:

this.path = resolve(this.ctx.baseDir, this.path)

Commit 6bb2963 ("test: fix tests") replaced that line with the URL form and, in the same commit, changed the sqlite test from join(__dirname, 'test.db') to new URL('test.db', import.meta.url).href. The test was adapted to the new code, so the break was not caught.

The same idiom was fixed twice recently in cordiverse/cordis by the same maintainer: packages/include/src/index.ts (merged cordiverse/cordis#200) and packages/hmr/src/index.ts (merged cordiverse/cordis#208), both with the comment "new URL() would read its C: as a URL scheme, which fileURLToPath rejects". This patch applies the same conversion (via pathToFileURL) to the sqlite driver. Relative paths, file: URLs, POSIX-absolute paths and UNC paths still go through new URL() unchanged.

The test case is in the existing packages/sqlite/tests/index.spec.ts, so the package keeps running as a single vitest worker. It asserts the resolved path, is gated to Windows, and a control case covers relative paths. On an unmodified main the new case fails with the TypeError above; with the patch the package suite is 269 passed (267 + 2). Locally the shared performance beforeAll needs more than vitest's default 10 s hook timeout, on main as well as with this patch.

Relative paths containing # or ? are also truncated by new URL(); that is a separate issue and is not changed here.