fix: emit Postgres column meta from result.fields (#11)

[ChuTingzj]

Fixes #11 — #11

Summary

Successful Postgres executeSql returned rows with empty columns, so hosts that zip columns[i] onto values[i] produced {} objects.

The mistake

The runner listened for a fields event on pg.Query. node-postgres 8.x never emits fields. Field metadata is on result.fields when row / end fire.

Stdout was:

{"type":"row","values":[...]}
...
{"type":"end","truncated":false}

with no {"type":"meta","columns":[...]}. Host consumeNdjsonResult therefore got columns: [].

MySQL is unchanged: mysql2 still emits fields, and streamMysql still uses that event.

The fix

sandbox/pg-query-stream.js emits meta from result.fields on the first row and/or end (at most once). Zero-row queries that still have a schema get columns on end. A fields listener remains as a no-op fallback for emitters that do fire it.

Tests

Regression coverage uses a mock Query that never emits fields and only provides result.fields:

• rows present → meta then rows then end
• zero-row result with a schema → meta then end
• row + end both carry fields → a single meta
• legacy fields event still works without duplicating meta
• MySQL path in execute.js still listens for fields

Runner image / npm

publish-runner.yml publishes on push to main (sandbox/**) or workflow_dispatch. This environment cannot dispatch that workflow (GitHub App 403: Resource not accessible by integration), and this PR is not merged, so GHCR :v1 is not rebuilt from these commits yet.

• DEFAULT_SANDBOX_IMAGE still pins the current digest until the Publish runner image job summary is available after merge (or a manual workflow_dispatch of this branch).
• npm is not published or tagged here. After the new digest lands, bump to 0.1.4 (the package embeds the digest) and follow the existing tag/publish-npm.yml flow — same two-step as ci: publish sandbox runner image to GHCR #2 → feat: pin DEFAULT_SANDBOX_IMAGE to published GHCR digest #3.

Test plan

• pnpm typecheck / pnpm test / pnpm build locally (58 tests, 0 failures)
• GitHub Actions CI test job green on this PR
• After merge (or workflow_dispatch of this branch): copy the new digest from the Publish runner image job summary into DEFAULT_SANDBOX_IMAGE + README
• Then bump package.json to 0.1.4 and tag v0.1.4 when ready to publish npm

  

[ChuTingzj]

Verdict

Fixes #11 correctly. Root cause (listening for non-existent pg 8.x fields event) is real; result.fields on row/end + single meta is the right fix. Regression tests force the no-fields Query shape. Dockerfile includes the new module. No must-fix in code.

Follow-up (ops, not this PR)

GHCR digest / DEFAULT_SANDBOX_IMAGE / npm 0.1.4 still required after merge — consumers on the current digest keep empty Postgres columns until then.

Do not merge from this bot.

[ChuTingzj]

Fix matches #11. pg 8.x Query does not emit fields; reading result.fields on row/end with metaSent once is the right shape. Zero-row + legacy fields covered. MySQL left on fields event — good.

Order is correct for the host consumer: meta is written before the first row event.

[ChuTingzj]

Required for the fix to ship in GHCR — without this COPY the container would still run the old inlined path (or fail to resolve ./pg-query-stream.js). Good that it is in the same PR.

Reminder (not a code block): after merge, Publish runner → new digest → bump DEFAULT_SANDBOX_IMAGE + npm 0.1.4. Old digest stays broken for Postgres columns until then.

[cursor]

Approved: Cursor Bugbot completed successfully with no findings that require human review. No reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}