Fix diagnose_mismatched_decls segfault

[Jan Hubicka]

> Jan Hubicka <jh@suse.cz> writes:
> 
> > We probably should have some testcases for the warnings comming out of
> > this magic function :)
> 
> We have many of these already.  If you would like to write more, be my
> guest, but please check for existing cases first.

I was mostly about to go sleep so wanted to note it before forgetting :)
I will write at least the inlining test, this is kind of mine area
anyway.
> 
> > What happens to alwaysinline?
> 
> Doesn't appear to be handled; probably should be (the combination of
> always- and no-inline is obviously silly).

On related note.  Currently
__attribute__ ((alwaysinline)) t();
does not make t inlined, one needs to do:
inline __attribute__ ((alwaysinline)) t();
would it be OK to add code setting DECL_INLINE/DECL_DECLARED_INLINE
in code setting alwaysinline attribute?
or shall the first case be warned on/erroed out?  (this sounds bit too
purist)

Honza
> 
> zw