From: Tom Lane <tgl@sss.pgh.pa.us>
To: pgsql-hackers@lists.postgresql.org
cc: Michael Meskes <meskes@postgresql.org>
Subject: Inconsistent function definitions in ECPG's informix.c
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2401574.1611764534.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Wed, 27 Jan 2021 11:22:14 -0500
Message-ID: <2401575.1611764534@sss.pgh.pa.us>
Precedence: bulk

I noticed that some of the newer compilers in the buildfarm
(e.g., caiman, with gcc 11.0) whine about the definitions of
rjulmdy() and rmdyjul() not quite matching their external
declarations:

informix.c:516:23: warning: argument 2 of type `short int[3]' with mismatc=
hed bound [-Warray-parameter=3D]
  516 | rjulmdy(date d, short mdy[3])
      |                 ~~~~~~^~~~~~
In file included from informix.c:10:
../include/ecpg_informix.h:38:31: note: previously declared as `short int =
*'
   38 | extern int      rjulmdy(date, short *);
      |                               ^~~~~~~
informix.c:567:15: warning: argument 1 of type `short int[3]' with mismatc=
hed bound [-Warray-parameter=3D]
  567 | rmdyjul(short mdy[3], date * d)
      |         ~~~~~~^~~~~~
In file included from informix.c:10:
../include/ecpg_informix.h:41:25: note: previously declared as `short int =
*'
   41 | extern int      rmdyjul(short *, date *);
      |                         ^~~~~~~

This isn't a bug really, since per the C spec these declarations
are equivalent.  But it'd be good to silence the warning before
it gets any more common.

The most conservative thing to do would be to take the user-visible
extern declarations as being authoritative, and change informix.c
to match.  I'm slightly tempted to do the opposite though, on the
grounds that showing the expected lengths of the arrays is useful.
But I wonder if anyone's compatibility checker tools would
(mistakenly) classify that as an ABI break.

Thoughts?

			regards, tom lane