pgbench: Fix error reporting in readCommandResponse().
authorFujii Masao <[email protected]>
Tue, 30 Sep 2025 14:52:28 +0000 (23:52 +0900)
committerFujii Masao <[email protected]>
Tue, 30 Sep 2025 14:53:46 +0000 (23:53 +0900)
commita912118c6055f683d51763546ee3c14b6e732893
tree25e47994bc0cba0ab11ebfb5504332d0c4fddd74
parent6778fbca63c9efffefee8086914cef8428c4ccad
pgbench: Fix error reporting in readCommandResponse().

pgbench uses readCommandResponse() to process server responses.
When readCommandResponse() encounters an error during a call to
PQgetResult() to fetch the current result, it attempts to report it
with an additional error message from PQerrorMessage(). However,
previously, this extra error message could be lost or become incorrect.

The cause was that after fetching the current result (and detecting
an error), readCommandResponse() called PQgetResult() again to
peek at the next result. This second call could overwrite the libpq
connection's error message before the original error was reported,
causing the error message retrieved from PQerrorMessage() to be
lost or overwritten.

This commit fixes the issue by updating readCommandResponse()
to use PQresultErrorMessage() instead of PQerrorMessage()
to retrieve the error message generated when the PQgetResult()
for the current result causes an error, ensuring the correct message
is reported.

Backpatch to all supported versions.

Author: Yugo Nagata <[email protected]>
Reviewed-by: Chao Li <[email protected]>
Reviewed-by: Fujii Masao <[email protected]>
Discussion: https://postgr.es/m/20250925110940.ebacc31725758ec47d5432c6@sraoss.co.jp
Backpatch-through: 13
src/bin/pgbench/pgbench.c