Re: RFC: Use -Wall -Wextra

[Eli Zaretskii <eliz@gnu.org>]

> Date: Fri, 29 Dec 2006 08:58:15 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: gdb-patches@sourceware.org
> > 
> > I would agree only if we never try to use -Werror, because with such
> > aggressive warnings GDB will never build if we add -Werror.
> > 
> > My other fear is that, with GCC becoming more and more picky about
> > perfectly valid C code, these options will cause the compilation to
> > become very noisy, but I guess we will hear complaints if that
> > happens.
> 
> I don't understand.  I built GDB with these warnings and -Werror, so
> that can't be literally true.

Well, it is on this machine:

  Linux fencepost 2.6.16.29-xen #1 SMP Wed Dec 6 07:32:36 EST 2006 x86_64 GNU/Linux

See below for the error message.

> I definitely don't want to back away from -Werror: I think it's done
> us a lot of good.

I'm not against -Werror, I just don't trust GCC to not flag some
innocent code as questionable under -Wall -Wextra.  These flags are
for those who intentionally want a noisy compiler; adding -Werror to
them is asking for too much trouble.

> I enabled -Wextra mostly because I wanted -Walways-true

IMO -Walways-true is one of the greatest nuisances with the latest GCC
versions.  Some code, especially sophisticated macros that need to
work portably with different data sizes, simply cannot be made to work
around this warning, especially if you want to handle 32-bit and
64-bit machines.  There already are such problems in GNU Make and in
Emacs, and developers are unwilling to fix them because it's too much
trouble, if at all possible.

> Can you give me some concrete examples of warnings you're concerned about?

Those issued by -Walways-true and the ones that wine about mismatch
between pointers to signed and unsigned char are the two that come to
mind.

> (Remember, we ship releases without -Werror.)

Thank God for small favors!

> > I'd advise against -Wunused: the problems it finds are harmless,
> > whereas fixing them is not trivial at all, and quite ugly in some
> > cases.
> 
> I'm afraid I don't understand this either.  The problems -Wunused finds
> are usually very easy to fix.

You mean with "i = i;"?

Here's the problem that prevented the latest snapshot from building
after I patched it with your patches:

    gcc -c -g -O2    -I. -I. -I./config -DLOCALEDIR="\"/usr/local/share/locale\"" -DHAVE_CONFIG_H -I./../include/opcode -I./../readline/.. -I../bfd -I./../bfd -I./../include   -DMI_OUT=1 -DTUI=1  -Wall -Wextra -Wpointer-arith -Wformat-nonliteral -Wno-pointer-sign -Wno-unused-parameter -Wno-unused -Wno-sign-compare -Wno-switch -Wno-missing-field-initializers -Werror infrun.c
    cc1: warnings being treated as errors
    infrun.c: In function 'handle_inferior_event':
    infrun.c:1458: warning: statement with no effect
    make[2]: *** [infrun.o] Error 1