Re: config headers and #undefs

[Kevin Ryde]

Akim Demaille <address@hidden> writes:
> 
> This is typically what should happen in config.h.  config.h is
> strictly equivalent to a series of -D flags passed to the compiler.
> Anything else, such as including a header, belongs to system.h.
> 
> Really, unless I misunderstood something, you are getting out of the
> Autoconf model.

This is for GMP, in the next release gmp.h will be generated at
configure time.  Initially this'll probably only have a define
controlling whether a "limb" is a long or a long long, but in the
future it'll get more settings, for instance whether certain library
functions exist or whether they should be macros.  In any case these
are basically aspects of libgmp.a determined at configure-time (based
on the host cpu, user choices, etc).

I thought of using AC_CONFIG_FILES, but that put settings through into
the makefiles, which is harmless but AC_CONFIG_HEADERS seemed better
suited to #defines.  I guess also there's no reason not to use a
separate installed file for the defines wanted, it was just that a
single file was so close to working.

By the way, is it intentional that the "pre" and "post" parts of an
AC_CONFIG_HEADERS get transformed?  If they got used literally that
would work too.