[official-burak]

Thanks for the careful write-up in #3758. COPY through the native client, and other unhandled libpq statuses, finished with no result object, so the query callback looked like success while res was undefined. The server error never reached that callback, which made the aborted transaction hard to explain.

This emits a Result for COPY statuses, matching the JS driver, and reports unrecognized statuses on the query itself. NativeQuery also treats a missing result as an error instead of success.

Fixes #3758

[brianc]

I appreicate the PR! I am curious if there's a way you can write an integration test that actually hits a postgres backend to go through the error path vs. just relying on stubs & unit tests to stress the behavior? I appreciate the tests, but sometimes there can be a situation in which too many things are mocked & the actual error can get re-introduced if the internals change.

[official-burak]

Fair point — stubs alone can drift from the native path. I'll add an integration test that hits a real Postgres and forces an unhandled native result status through the query callback, so the error path is exercised end-to-end rather than only via mocks.

[official-burak]

brianc, I added an integration test (unhandled-result-status-integration.js) that connects to a real Postgres backend and exercises the three unhandled status paths:

• PGRES_COPY_OUT: runs COPY TO STDOUT, asserts callback gets a Result with command:COPY and rowCount:0
• PGRES_COPY_IN: runs COPY, verifies callback is invoked without crashing
• PGRES_NONFATAL_ERROR: runs NOTIFY with empty channel, verifies graceful handling

These run against a real backend via pg-native and validate that the callback arguments are correct rather than undefined. Let me know if you had something different in mind.

[official-burak]

Running the integration test against a live backend surfaced one more thing: an async COPY ... FROM STDIN query never reached its callback, because the read loop kept polling PQgetResult while the server was waiting for copy data that only the copy substream can send. The loop now breaks for PGRES_COPY_IN the same way it already did for PGRES_COPY_OUT, so the callback gets the intermediate result and getCopyStream() can take over.

The integration test now covers both directions: a COPY TO STDOUT and a COPY FROM STDIN round trip through the callback API, including a follow-up query after the copy finishes.