[iamralch]

Problem

QueryResult::last_insert_rowid() returns Option<i64>, but sqlx-sqlite's SqliteQueryResult::last_insert_rowid() returns i64. Code written against sqlx-sqlite therefore doesn't compile on D1 or Durable Objects, even though the drivers otherwise follow sqlx-sqlite wherever JavaScript doesn't force a difference (sqlx-cloudflare-core/src/types.rs). JavaScript doesn't force this one.

This also affects sqlc-gen-sqlx's SQLite output: its :execlastid functions end in Ok(result.last_insert_rowid()) returning i64.

Change

• last_insert_rowid() returns i64, 0 when the statement inserted no row. QueryResult::new takes an i64 to match.
• D1 maps a missing meta.last_row_id to 0. DO passes last_insert_rowid() through, and a statement that wrote nothing reports 0, as before but no longer None.
• Extending a result still keeps the last non-zero rowid, so a trailing SELECT doesn't erase the INSERT before it.
• Worker test assertions and the DO README follow.

Rowids start at 1 unless a row is given 0 explicitly, so 0 is unambiguous in practice; RETURNING is the way to be certain. sqlx-sqlite reports a stale rowid after a SELECT, so 0 is the more accurate answer of the two.

Breaking

feat!: last_insert_rowid() returns i64, and 0 where it returned None. release-please should bump to 0.2.0.

Verification

• make check lint test passes, host and wasm32.
• Clippy on both test Workers for wasm32: only the "very complex type" warning that main already has.
• sqlc-gen-sqlx's SQLite example output (from the feat/engine-sqlite fork branch), with only the AsExecutor impls retargeted, compiles against both D1 and Do. The :execlastid function compiles unchanged.
• make test-worker not run locally; CI runs it.