[vishalhabib99]

Summary

The SQLite connector splits a multi-statement batch into writes and reads and runs all writes before any reads. A read that comes before a write in the batch therefore sees the write's effects:

SELECT a AS before_delete FROM t;   -- t has 3 rows
DELETE FROM t;
SELECT count(*) AS after_delete FROM t;

returns DELETE (3), then before_delete: [], then after_delete: 0. The first SELECT comes back empty and the result sets are reordered. PostgreSQL (and MySQL/MariaDB via native multi-statement) run in source order.

The split dates from when writes were batched through .exec() for speed (16e870b). Writes now run one at a time via prepare().run(), so it no longer buys anything. This PR runs each statement in the order written and branches on returnsRows() per statement. The per-statement PRAGMA query_only = ON re-assert and the maxRows truncation probe are unchanged.

I updated should handle maxRows in multi-statement execution, which asserted the writes-first order. If that order is intended, happy to close this and keep the behavior documented instead.

Test plan

• New should execute multi-statement batches in source order (fails on main, passes here)
• Updated should handle maxRows in multi-statement execution, which asserted the writes-first order
• pnpm test:unit: 1201/1201
• sqlite, multi-sqlite, execute-sql and json-rpc integration tests pass
• Read-only mode still rejects SELECT ...; PRAGMA query_only = OFF; DELETE ...

[Copilot]

🟢 Approval recommended

The implementation is focused, preserves existing safeguards, and includes appropriate integration coverage.

0 open findings

What changed in this PR

Runs SQLite multi-statement batches in source order, aligning results with statement order.

Changes:

• Executes each SQLite statement sequentially.
• Adds and updates integration coverage for ordering and row limits.

File	Description
src/​connectors/​sqlite/​index.ts	Preserves source execution order.
src/​connectors/​__tests__/​sqlite.integration.test.ts	Verifies ordering and updated semantics.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[tianzhou]

LGTM. Thanks for the contribution.

[vishalhabib99]

Thanks @tianzhou for the quick review and merge.