tursodatabase/turso - CodSpeed

turso

Blog Docs Changelog

Performance History

Latest Results

Update generated JS bindings

window-filter

1 hour ago

core/storage: add red-first coverage for the saveAllCursors port Two layers covering the runtime added in #7341: - Internals (storage::btree::tests::save_all_cursors, 8 cases) pin the individual primitives — peer registration, has_peers toggling, try_save_position_for_external_balance's Saved/MustInvalidate outcomes, restore_context's CURSOR_SKIPNEXT semantics, and clear_btree's auto-invalidation of peers. 4 of the 8 go red when drive_pending_peer_save and invalidate_peer_cursors are stubbed out; the remaining 4 cover the registry API surface that is exercised even on the fast path. - SQL surface (testing/sqltests/tests/save-all-cursors.sqltest, 4 cases) extends the window-selfjoin-reset-sorter pattern: a window function over a triple self-joined source forces ephemeral overflow + ResetSorter, leaving OpenDup peer cursors that need the registry's auto-invalidate. Variants cover multi-partition data, DELETE driven by a scalar subquery, UPDATE driven by an IN subquery, and a materialized-CTE source. All four fail with "parent page 2 is a leaf page, expected interior page" when the runtime is reverted.

save-all-cursors

1 hour ago

Add test-only helper for nondeterministic functions

window-filter

3 hours ago

prevent corrupting logical log frames by yielding mid read

PThorpe92:nonblock

6 hours ago

core: preserve original-case schema names; ASCII-only identifier folding Store table/view/index names in sqlite_schema with their original case (matching SQLite) instead of the lowercased form, while keeping name resolution case-insensitive. Fold identifiers ASCII-only (to_ascii_lowercase) so non-ASCII identifiers like `cœur` and `CŒUR` (or `é`/`É`) stay distinct, as in SQLite, and so databases containing such objects can be opened. DDL matching against sqlite_schema (DROP TABLE/VIEW/INDEX, ALTER RENAME, and the post-CREATE ParseSchema) compares names with NOCASE so it works for both new (original-case) and legacy (lowercased) rows. Databases written by a pre-fix Turso version persisted non-ASCII identifiers Unicode-lowercased (e.g. `CŒur` stored as `cœur` in sqlite_schema.name while the sql column kept the original case). On load we detect that mismatch and alias the original ASCII-folded name to the persisted key, so such tables stay addressable by their original case and DROP/ALTER remove them cleanly instead of orphaning the row. Covered by a committed old-binary fixture + regression test. Fixes #5781 Fixes #5730 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

glommer:fix/ident-case

9 hours ago

bindings/rust: Add multiprocess WAL builder test Add a Rust binding test that keeps one process connected with multiprocess WAL enabled, then checks that a second process can open the same database and write to it. This proves the second-process open issue is fixed. Reviewed-by: Triston Armstrong <Tarmstrong95@proton.me> Reviewed-by: Lucas Montes <lluc23@hotmail.com>

davidabram:fix-experimental_multiprocess_wal-wal-file

9 hours ago

Merge 'do not rollback attached pagers on reprepare if the conn is inside transaction' from Pedro Muniz The problem here is that you could have the statement error with database changed after creating the temp table inside a transaction, or creating it outside of the transaction but then trying to insert into into it, but not seeing the inserted row. ``` 1. CREATE TEMP TABLE inside BEGIN, followed by INSERT/SELECT, returns "Database schema changed". 2. CREATE TEMP TABLE outside BEGIN works, but INSERT into that temp table inside BEGIN returns success while the row is not visible inside the transaction or after COMMIT. ``` This is my attempt to fix this after comparing the behaviour with sqlite. Essentially we should not rollback the attached pagers if we are in explicit transaction, and seems like we should not rollback at all for temp tables (maybe I'm wrong on this appreciate some rebuttal). Fix temp tables with reprepare AI found the potential issue. Reviewed-by: Preston Thorpe (@PThorpe92) Closes #7348

main

9 hours ago

migrate wal.rs to use fallible vec allocations

pedrocarlo:wal-alloc

10 hours ago

Latest Branches

0%

core/translate: support FILTER in window clauses#7265

2 hours ago

a83f907

window-filter

0%

core/storage: port SQLite's saveAllCursors to a pager-level registry#7341

2 hours ago

80c5c49

save-all-cursors

0%

Remove blocking IO from mvcc bootstrap and recovery paths#7349

7 hours ago

d7ebacb

PThorpe92:nonblock

© 2026 CodSpeed Technology

Home Terms Privacy Docs