Okay, pushed this patch and the new testing for it based on libpq_pipeline. We'll see how the buildfarm likes it.
I made some further changes to the last version; user-visibly, I split the trace flags in two, keeping the timestamp suppression separate from the redacting feature for regression testing. I didn't like the idea of silently skipping the redacted fields, so I changed the code to print NNNN or SSSS instead. I also made the redacting occur in the last mile (pqTraceOutputInt32 / String) rather that in their callers: it seemed quite odd to advance the cursor in the "else" branch. I refactored the duplicate code that appeared for Notice and Error. In that function, we redact not only the 'L' field (what Iwata-san was doing) but also 'F' (file) and 'R' (routine) because those things can move around for reasons that are not interesting to testing this code. In the libpq_pipeline commit I added 'pipeline_abort' and 'transaction' to the cases that generate traces, which adds coverage for NoticeResponse and ErrorResponse. -- Álvaro Herrera 39°49'30"S 73°17'W
