[fireboltdude1357]

The fork's nightly build fails because patch/sqlite-reader-pool conflicts with upstream b23f94e in packages/shared/src/nodeSqliteClient.ts (6 hunks).

Upstream pingdotgg#35728 moves the writer onto a worker thread and rewrites the file around a shared makeWorkerConnection. It still routes reads by claim count over 4 readers, each with its own Semaphore(1), so a new read can overtake one that is already waiting.

I kept upstream's makeWorkerConnection, makeReader, writer worker, checkpoint worker and READER_POOL_SIZE exactly as they are. The patch's ordered pool is back as makeReaderPool(filename), built over upstream's makeReader:

• the lowest free slot goes out first;
• waiters are served in arrival order;
• when the readers turn off, queued reads move to the writer in order;
• a slot taken by an interrupted caller goes straight to the next waiter.

route's reader branch now leases a slot from that pool and falls back to upstream's writer. Readers no longer take a per-reader lock, because the slot already gives them the connection alone. db.lock_wait_ms for a reader counts the time spent waiting for a slot.

nodeSqliteClient.test.ts passes, with 27 tests including the patch's 4 pool tests. pingdotgg#35728's Sqlite.test.ts, FoundationPersistence.test.ts and ProjectService.test.ts pass too.

On a scratch stack of upstream b23f94e with all 26 patches and this branch, pnpm install --no-frozen-lockfile, vp run typecheck (41 of 41) and vp test run on the test files the patches touch, plus those upstream tests, all pass (51 files, 1439 tests).

Made by Claude Opus 5.5 in Claude Code.

🤖 Generated with Claude Code