This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: retry_ice mechanism doesn't work reliably with C++
- From: Joe Buck <Joe dot Buck at synopsys dot COM>
- To: Michael Matz <matz at suse dot de>
- Cc: Jakub Jelinek <jakub at redhat dot com>, gcc at gcc dot gnu dot org
- Date: Mon, 1 Mar 2004 10:22:06 -0800
- Subject: Re: retry_ice mechanism doesn't work reliably with C++
- References: <Pine.LNX.4.58.0403011531540.24101@wotan.suse.de>
On Mon, Mar 01, 2004 at 03:39:07PM +0100, Michael Matz wrote:
> Hi Jakub,
>
> we just noticed that sometimes the retry_ice doesn't find some ICEs
> reproducable with C++. The problem being that it compares the assembler
> output of the different tries, and bails out if they are different. For
> anon namespaces the mangled symbols are randomized making the output
> unequal, leading to the usual "not reproducible" warning.
And this (mangling anon namespace symbols instead of making them static)
is a feature that I would dearly like to get rid of, or at least use more
rarely. The reasons given for needing to make anonymous namespace symbols
global apply to rather theoretical cases, mainly involving the "export"
feature which we have not even implemented. It seems an overly cautious
approach that creates more problems than it solves.