Re: Automatic PG_PRINTF_ATTRIBUTE

On Fri, Nov 21, 2014 at 01:16:25PM -0500, Tom Lane wrote:
> Noah Misch <noah(at)leadboat(dot)com> writes:
> > pg_config_manual.h has been choosing gnu_printf as the PG_PRINTF_ATTRIBUTE for
> > every MinGW build. That invites a torrent of warnings on pre-gcc-4.4 MinGW
> > compilers, including the compiler on buildfarm member narwhal. I'm
> > increasingly using an affected compiler, because it builds twice as quickly as
> > today's gcc. Let's have "configure" detect whether gcc supports gnu_printf
> > before using it. I gather plain "printf" aliases ms_printf on Windows and
> > gnu_printf elsewhere. Therefore, while the new "configure" test applies to
> > all platforms, non-Windows platforms are disinterested in the outcome today.
> > Suppose gcc introduces aix_printf and has plain "printf" alias it on AIX.
> > PostgreSQL will continue to replace platform printf implementations that
> > depart from our format processing expectations, and our own elog.c code
> > processes errmsg() formats. Therefore, gnu_printf would remain the better
> > global choice even if new archetypes become available.
>
> No objection here. I'm not 100% convinced by your argument that we'd not
> need to modify the logic in future ... but if we do, we'd still be better
> off having it in a configure test than trying to get an #ifdef nest to
> do the right thing.

Agreed. The key take-away is that I opted to use gnu_printf on all compilers
supporting it, not just on Windows compilers supporting it.

> What about the MSVC build path? I guess there we're only targeting
> one compiler, so it should be easy.

c.h zaps __attribute__(...) under non-__GNUC__ compilers, so we need not touch
the MSVC build system.