[patch]: Fix structure va_list type treating in std_canonical_va_list

[Bob Wilson]

Kai Tietz wrote:
> "Richard Guenther" <richard.guenther@gmail.com> wrote on 14.07.2008 
> 10:43:43:
>> How did you test this patch?
> 
> I did a test for w64 and for cross x86_64-pc-linux64. But this regression 
> test is somehow meaningless, because this function isn't used for those 
> targets. I have no xtensa environment. I hope that Bob could test this 
> patch, because he suggested a patch (see 
> http://gcc.gnu.org/ml/gcc-patches/2008-07/msg00873.html), which leads to 
> the described problems - I mentioned above. It is obvious (for me) that 
> the type check of va_arg is broken by this approach, because it is based 
> on the fact, that canonical_va_list should return for invalid types a 
> NULL_TREE (as you suggested while the initial patch review).

It works for Xtensa.  Sorry that it took me all day to test.  There are a number 
of testsuite failures for Xtensa right now, so I compared your patch against my 
temporary workaround to make sure that none of the failures were related to 
this.  It takes a while to run the testsuite twice with a simulator!