[LucaCappelletti94]

Part of #202. Stacked on #205, so only 2172557 is added here.

memvfs keeps real SQLite lock levels per file, following the unix VFS, so connections sharing a database get SQLITE_BUSY on conflict. Two connections to one memvfs database used to share pages without any locking, which the docs called unsupported. This is the only change for builds without threadsafe.

With threadsafe, which now also turns on rsqlite-vfs/threadsafe, the file map and every file are guarded by SQLITE_MUTEX_FAST mutexes from SQLite's mutex API and shared through Arc, so those connections may live on different workers. install then requires the OsCallback to be Send + Sync, expressed by a new MemVfsOs trait that every OsCallback + 'static satisfies without the feature. The last VFS error becomes per thread, since xGetLastError names no connection, which pulls in std.

[Spxg]

Thanks for the PR. But how does it work correctly on wasm? With SQLITE_THREADSAFE=0, I don't think the mutex-related symbols would be exported.

[Spxg]

The feature needs a SQLite built with SQLITE_THREADSAFE above 0, so sqlite-wasm-rs turns it on only in the follow-up that adds its own threadsafe feature.

Oh, I see. However, I'm curious whether it's possible to have a proper mutex implementation. Did you use futex-based mutexes or no-op mutexes in your tests?

[LucaCappelletti94]

Oh, I see. However, I'm curious whether it's possible to have a proper mutex implementation. Did you use futex-based mutexes or no-op mutexes in your tests?

Futex-based, never no-op. The native tests here use libsqlite3-sys's bundled SQLite, built with SQLITE_THREADSAFE=1 and pthread mutexes.

For Wasm, the follow-up PR adds a futex-style mutex on memory.atomic.wait32 and makes it SQLite's default. Its worker tests fail with the no-op mutexes. It is hand-written because std and parking_lot sleep in memory.atomic.wait32 under contention without checking the thread, which traps on Chrome's main thread, and SQLite needs a recursive mutex that it enters and leaves in separate C calls.

I will open the stacked PR shortly so you can have a look, I have been trying to find either some seams or possible cleanups to make it smaller as for now it is ~1k lines.