max_align_t mess

max_align_t mess

[Rich Felker]

Sometime we need to decide what to do about max_align_t. The GCC
definition is:

typedef struct {
  long long __max_align_ll __attribute__((__aligned__(__alignof__(long long))));
  long double __max_align_ld __attribute__((__aligned__(__alignof__(long double))));
} max_align_t;

Unfortunately, this definition is relying on bugs in GCC to give the
"right" result (the result they want and that matches the existing
ABIs), and is not reproducible in any portable manner: they're relying
on __alignof__(T) to give the preferred alignment of T rather than the
required alignment of T, and the portable (C11) _Alignof will not give
this result except on old buggy GCC.

One sane way to handle this might be to just put max_align_t in the
arch-specific alltypes.h fragment and use the following definition:

TYPEDEF struct { _Alignas(N) struct { char __align_c[M]; }; } max_align_t;

where N and M are the correct alignment and size for the arch ABI.

Rich

Re: max_align_t mess

[Szabolcs Nagy]

* Rich Felker <dalias@libc.org> [2014-06-23 00:48:07 -0400]:
> 
> TYPEDEF struct { _Alignas(N) struct { char __align_c[M]; }; } max_align_t;
> 
> where N and M are the correct alignment and size for the arch ABI.
> 

I like this but unfortunately it only works if the compiler has _Alignas

Re: max_align_t mess

[Rich Felker]

On Mon, Jun 23, 2014 at 09:51:35AM +0200, Szabolcs Nagy wrote:
> * Rich Felker <dalias@libc.org> [2014-06-23 00:48:07 -0400]:
> > 
> > TYPEDEF struct { _Alignas(N) struct { char __align_c[M]; }; } max_align_t;
> > 
> > where N and M are the correct alignment and size for the arch ABI.
> > 
> 
> I like this but unfortunately it only works if the compiler has _Alignas

Since max_align_t can only be defined portably with C11, I think it's
reasonable to expose it only under __STDC_VERSION__ >= 201112L. Or we
could add an #elif __GNUC__ using __attribute__((__aligned__)) but
from a namespace standpoint it's probably preferable to hide it for
pre-C11 anyway.

Rich