On Thu, Jul 10, 2014 at 11:10 PM, Sawada Masahiko <sawada.mshk@gmail.com> wrote:
> Hi,
>
> At 1047 line of receivelog.c:CopyStreamPoll(), we set NULL to
> timeoutptr variable.
> if the value of timeoutprt is set NULL then the process will wait
> until can read socket using by select() function as following.
>
> if (timeout_ms < 0)
> timeoutptr = NULL;
> else
> {
> timeout.tv_sec = timeout_ms / 1000L;
> timeout.tv_usec = (timeout_ms % 1000L) * 1000L;
> timeoutptr = &timeout;
> }
>
> ret = select(PQsocket(conn) + 1, &input_mask, NULL, NULL, timeoutptr);
>
> But the 1047 line of receivelog.c is never executed because the value
> of timeout_ms is NOT allowed less than 0 at CopyStreamReceive which is
> only one function calls CopyStreamPoll().
> The currently code, if we specify -s to 0 then CopyStreamPoll()
> function is never called.
> And the pg_receivexlog will be execute PQgetCopyData() and failed, in
> succession.
Thanks for reporting this! Yep, this is a problem.
> I think that it is contradiction, and should execute select() function
> with NULL of fourth argument.
> the attached patch allows to execute select() with NULL, i.g.,
> pg_receivexlog.c will wait until can read socket without timeout, if
> -s is specified to 0.
Your patch changed the code so that CopyStreamPoll is called even
when the timeout is 0. I don't agree with this change because the
timeout=0 basically means that the caller doesn't request to block and
there is no need to call CopyStreamPoll in this case. So I'm thinking to
apply the attached patch. Thought?
Regards,
--
Fujii Masao
