Re: perl 5.36, C99, -Wdeclaration-after-statement -Wshadow=compatible-local

Andres Freund <andres@anarazel.de>

From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Peter Eisentraut <peter.eisentraut@enterprisedb.com>, pgsql-hackers@postgresql.org, David Rowley <dgrowleyml@gmail.com>, Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
Date: 2022-12-29T18:42:36Z
Lists: pgsql-hackers

Attachments

Hi,

On 2022-12-28 19:05:35 -0500, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > On 2022-12-28 13:43:27 -0500, Tom Lane wrote:
> >> Hmm ... I guess the buildfarm would tell us whether that detection works
> >> correctly on platforms where it matters.  Let's keep it simple if we
> >> can.
> 
> > Quick clarification question: Are you suggesting to use #ifdef __GNUC__, or
> > that it suffices to use -Werror=unknown-pragmas without a separate configure
> > check?
> 
> I'd try -Werror=unknown-pragmas, and then go to the #ifdef if that
> doesn't seem to work well.

It turns out to not work terribly well. gcc, quite reasonably, warns about the
pragma used in .c files, and there's no easy way that I found to have autoconf
name its test .h. We could include a test header in the compile test, but that
also adds some complication. As gcc has supported the pragma since 2000, I
think a simple
  #ifdef __GNUC__
  #define HAVE_PRAGMA_SYSTEM_HEADER	1
  #endif
should suffice.


I started to wonder if what we should do instead is to do something like

#ifdef HAVE_PRAGMA_GCC_DIAGNOSTIC
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wdeclaration-after-statement"
#pragma GCC diagnostic ignored "-Wshadow=compatible-local"
#endif

#include "EXTERN.h"
#include "perl.h"

#ifdef HAVE_PRAGMA_GCC_DIAGNOSTIC
#pragma GCC diagnostic pop
#endif

but that ends up quite complicated because gcc will warn about unknown
warnings being ignored:

../../../../home/andres/src/postgresql/src/pl/plperl/plperl.h:87:32: warning: unknown option after ‘#pragma GCC diagnostic’ kind [-Wpragmas]
   87 | #pragma GCC diagnostic ignored "-Wfrakbar"

which would mean we'd need to define a pg_config.h symbol for each potentially
ignored warning, and to guard each '#pragma GCC diagnostic ignored "..."' with
an #ifdef.


Thus I propose the attached.


Should we backpatch this? Given the volume of warnings it's probably a good
idea. But I'd let it step in HEAD for a few days of buildfarm coverage first.

Greetings,

Andres Freund

Commits

  1. perl: Hide warnings inside perl.h when using gcc compatible compiler