Re: [patch i386] Fix detection of va_list type for w64.

[Uros Bizjak <ubizjak at gmail dot com>]

Kai Tietz wrote:
> Hi Uros,
>
> 2008/7/7 Uros Bizjak <ubizjak@gmail.com>:
>   
> > Hello!
> >
> >     
> > > 2008-07-07  Kai Tietz  <kai.tietz@onevision.com>
> > >
> > >        * config/i386.c (is_va_list_pointer_char): New.
> > >        (ix86_va_start): Replace compare with ms_va_list_type_node
> > >        by call of is_va_list_pointer_char.
> > >        (ix86_gimplify_va_arg): Likewise.
> > >
> > > Bootstraped on x86_64-pc-linux64 and x86_64-pc-mingw32 without
> > > regressions.
> > > Ok for apply?
> > >
> > >       
> > Please add something like following to both call sites:
> >
> > if (!TARGET_64BIT
> >   || ix86_canonical_va_list_type (TREE_TYPE (valist)) ==
> >   (DEFAULT_ABI == MS_ABI) ? ms_va_list_type_node : va_list_type_node)
> > {
> >     
>
> This would be wrong, because - as I wrote - both return values are
> possible for w64. This is depent on the use of the
> __builtin_ms_va_list or __builtin_va_list, and on the use in context
> of x86_64 or in w64 itself. Your suggested replacement just checks for
> one case, which is just the half of the job here.
> The method checks for both types, which is the fix.
>   

Thanks for the explanation! Please go ahead with your original patch 
then. Also, please fix the comment before is_va_list_pointer_char. 
Return value is obvious from the function name ;)

Thanks,
Uros.