This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Fix PR optimization/12085
- From: Roger Sayle <roger at eyesopen dot com>
- To: Eric Botcazou <ebotcazou at libertysurf dot fr>
- Cc: Gabriel Dos Reis <gdr at integrable-solutions dot net>, <gcc-patches at gcc dot gnu dot org>
- Date: Sun, 7 Dec 2003 08:08:25 -0700 (MST)
- Subject: Re: [PATCH] Fix PR optimization/12085
On Sun, 7 Dec 2003, Eric Botcazou wrote:
> 2003-12-07 Eric Botcazou <ebotcazou@libertysurf.fr>
>
> PR optimization/12085
> * tree-inline.c (expand_call_inline): Do not inline functions at
> calling points where they are viewed with too different a prototype
> than the actual one.
>
> * gcc.dg/inline-5.c: New test.
>
This is OK for mainline with one change. I think that you should remove
the aborts inside the "#ifdef ENABLE_CHECKING" and always just return
NULL_TREE. There's no point ICEing the compiler in the middle-end just
because the user has entered some dubious code. Especially, when there's
a safe recovery strategy, i.e. just don't inline the function. It might
be reasonable to issue a warning instead, but even that task is best left
to the front-ends, and if the language doesn't think this sort of thing
merits a diagnostic, the middle-end tree inliner should remain silent on
the subject.
It's good that GCC survives an enable-checking bootstrap with the aborts
still present, but without BOOT_CFLAGS="-O3" we're not really testing as
much of the compiler as we could. For extra credit, you might try a
bootstrap of the patched GCC at -O3 before you remove the aborts, just
to confirm we're not doing anything dubious in the compiler itself, as
a one-time check.
This patch (with the above change) is clearly a safe bug-fix, and until
very recently the PR was even targeted for gcc 3.3.3, so I believe that
its suitable for both mainline and the gcc-3_3-branch. Gabriel?
Many thanks again.
Roger
--