On 09/04/2009 03:17 PM, Tobias Burnus wrote:
> On 09/04/2009 02:52 PM, Kai Tietz wrote:
>> Well, I bootstrapped this with Paolo's stdint.m4 patch, and it
>> workedon linux32 and linux64 (i386), and for x86_64-pc-mingw32, too.
>> Just out of curiosity, who is the size in gfortran frontend for
>> int_{least,fast}*_t done, as for gcc until now no stdint.h/inttypes.h
>> file was used?
>
> Actually, I do not know how exactly it works ... I now checked - and
> here is how it works:
>
> In gcc/fortran/ one calls:
>     TYPE_PRECISION (type)
> for INT_FAST64_TYPE etc.  The defaults are set in
>    gcc/default.h
>
> For instance:
>    #ifndef INT_FAST64_TYPE
>    #define INT_FAST64_TYPE ((const char *) NULL)
>    #endif
>
> And the INT_*_TYPE definitions can seemingly found in
> config/*/*stdint.h. For instance:
> ./i386/cygwin-stdint.h:#define INT_FAST64_TYPE "long long int"
> ./i386/mingw-stdint.h:#define INT_FAST64_TYPE "long long int"
> ./glibc-stdint.h:#define INT_FAST64_TYPE (LONG_TYPE_SIZE == 64 ? "long
> int" : "long long int")
>
> Thus, stdint.h is not directly involved and the patch should have no
> (adverse) effect on gfortran's c_int_fast*_t support.

Yes, the stdint.h we're talking about is for the host.  libgfortran 
(where the GCC target is actually the host, as in all target libraries) 
does not use int_fast and int_least.  Only the front-end itself cares, 
but it uses info from the target configuration headers.

Actually I'm pretty sure that stdint.h is not needed in the GCC tree 
only the intNN_t types (managed by GCC_STDINT_TYPES.  But since 
GCC_HEADER_STDINT is used by bfd too and I didn't want Kai to mess up 
with bfd, I provided it in my stdint.m4 rewrite.

Here is the latest and greatest version.

Paolo