Loading
Commits on Source 4
-
Ian Chechin authored
MAX/MIN/COALESCE over a DATETIME TEXT column drop the declared type (sqlite3_column_decltype returns "" for aggregates), so Next delivered a raw string and a Scan into *time.Time failed with "unsupported Scan, storing driver.Value type string into type *time.Time". The column-metadata route the reporter suggested does not help: on master sqlite3_column_origin_name already resolves direct columns (ColumnInfo, !113) but returns "" through MAX/MIN/COALESCE, so it cannot recover the source type. This does the runtime best-effort parse instead: an empty-decltype TEXT column under _texttotime is parsed via parseTime; success yields time.Time, failure falls back to the original string, so no Scan that worked before can newly fail. Gated on _texttotime so non-opt-in callers are unaffected. ColumnTypeScanType is left reporting string for empty decltype, since it is called before any row and cannot know an aggregate is date-shaped; Next ...
-
Ian Chechin authored
Addresses !133 review: reword the rows.go comment to cover every empty-decltype TEXT case (aggregates/expressions, subqueries, typeless real columns) rather than only MAX/MIN/COALESCE, document the *string RFC3339Nano reformat consequence on the _texttotime DSN doc, and add the v1.54.0 CHANGELOG entry.
-
cznic authored
Re-vendor lib/ and vec/ from libsqlite3 v1.14.2 and libsqlite_vec v0.3.1 (SQLite 3.53.3) and bump the pinned modernc.org/libc to v1.74.1, the version the transpile was generated against. Updates doc.go, CLAUDE.md, and the pending v1.54.0 CHANGELOG section. make vendor build_all_targets passes; local test suite (root incl. TestVec, pcache, vfs) is green. Co-Authored-By:Claude Opus 4.8 (1M context) <noreply@anthropic.com>