On 2022-Jul-04, Kyotaro Horiguchi wrote:
> At Wed, 29 Jun 2022 14:09:17 +0200, Alvaro Herrera <alvherre@alvh.no-ip.org> wrote in
> > However, another problem case, not fixed by PIPELINE_IDLE, occurs if you
> > exit pipeline mode after PQsendQuery() and then immediately use
> > PQexec(). The CloseComplete will be received at the wrong time, and a
> > notice is emitted nevertheless.
>
> Mmm. My patch moves the point of failure of the scenario a bit but
> still a little short. However, as my understanding, it seems like the
> task of the PQpipelineSync()-PQgetResult() pair to consume the
> CloseComplete. If Iinserted PQpipelineSync() just after PQsendQuery()
> and called PQgetResult() for PGRES_PIPELINE_SYNC before
> PQexitPipelineMode(), the out-of-sync CloseComplete is not seen in the
> scenario. But if it is right, I'd like to complain about the
> obscure-but-stiff protocol of pipleline mode..
Yeah, if you introduce PQpipelineSync then I think it'll work okay, but
my point here was to make it work without requiring that; that's why I
wrote the test to use PQsendFlushRequest instead.
BTW I patch for the problem with uniqviol also (not fixed by v7). I'll
send an updated patch in a little while.
> > I produced pipeline_idle.trace file by running the test in the fully
>
> By the perl script doesn't produce the trace file since the list in
> $cmptrace line doesn't contain pipleline_idle..
Ouch, of course, thanks for noticing.
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/