> On 14 Sep 2022, at 17:06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Daniel Gustafsson <dgustafsson@postgresql.org> writes:
>> Since pg_recvlogical is also supposed to run as a daemon, teach it about
>> SIGTERM as well and update the documentation to match. While in there,
>> change pg_receivewal's time_to_stop to be sig_atomic_t like it is in
>> pg_recvlogical.
>
> While looking at this commit, I wondered why both of those programs
> are declaring their signal handlers like
>
> static void
> sigint_handler(int signum)
>
> rather than our standard convention
>
> static void
> pmdie(SIGNAL_ARGS)
Sorry, I had missed that convention.
> Evidently we don't (any longer?) have any platforms where SIGNAL_ARGS
> is non-default, but that still doesn't make this good coding. More
> than once I've grepped for SIGNAL_ARGS to locate signal handlers.
>
> Hmm, looks like same mistake in pg_waldump ...
pg_test_fsync seems to have the same error.
> barring objection,
> I'm going to run around and fix those.
No objections, unless you have better things to do and wan't me to clean up
then I'll take care of it.
> I also notice that port.h has
>
> typedef void (*pqsigfunc) (int signo);
> extern pqsigfunc pqsignal(int signo, pqsigfunc func);
>
> which is a bit inconsistent with the idea that SIGNAL_ARGS
> might be different from that. Shouldn't we declare it as
>
> typedef void (*pqsigfunc) (SIGNAL_ARGS);
Agreed, that makes sense.
--
Daniel Gustafsson https://vmware.com/