[Bug-gnulib] Re: gnulib module for uintmax_t? (and best practice questions)

[Simon Josefsson]

Bruno Haible <address@hidden> writes:

> Paul Eggert wrote:
>> "configure" could put its stdint.h additions into config.h as
>> macros.  ...  This approach is simpler and I think more reliable
>
> Indeed this is much more robust, but to get everything defined for
> emulating a full stdint.h, you often not only have to #include the system's
> <stdint.h>, but also <stddef.h>, <limits.h>, <sys/inttypes.h> or <inttypes.h>.
>
> The config.h would then have to include system header files. That's a thing
> that I absolutely try to avoid, 

I think I agree with that goal, however some comments:

> because
>   1) on AIX 3, no other system include files are allowed before our 
> <alloca.h>,

config.h could include alloca.h too, and do it first.

>   2) with GCC up to ca. 3.0, you get an error if you make a global register
>      declaration after having #included some system header files which happen
>      to define inline functions. (Global register declarations are used by
>      GNU clisp.)

Can't that simply be considered a GCC bug, then?

Still, this does not solve the uint32_t-in-installed-headers problem.

Thanks.