Re: emacs-26 9e59de9: Use GCALIGNED properly for GCC

[Paul Eggert]

Stefan Monnier wrote:
> >    union gcaligned_foo { struct foo s; int64_t gcaligned; };
>
> Are int64_t necessarily aligned on multiples of 8 on 32bit platforms?

No, unfortunately.

> > IOW, should we rely on alignas?  There could be dragons there too, no?
>
> FWIW, for the dummy alignment thingy I wouldn't use `char` (I wouldn't be
> surprised to see errors in compilers when asking to align on multiples
> of N for objects smaller than N), so maybe
>
>      #define gc_aligned(typename) \
>         union { typename s; int64_t alignas (GCALIGNMENT) dummy; };

That does not work either, alas, as C11 says 'alignas (8)' is an error when the 
natural alignment of the object is less than 8. This is one of the problems that 
we have encountered in earlier attempts to fix this bug. 'char alignas (8)' 
avoids this problem.

If we run into a platform where alignas (8) does not work either natively or via 
Gnulib emulation, the patch I'm working on has a 'verify' check that should 
result in a build failure. Although I have some ideas for fixing the situation 
if it arises, they would add some complexity (and would depend on the details of 
any problematic hosts), and I'd rather avoid this if possible.