[RFC] Fortran search path for intrinsic includes and modules

[Jakub Jelinek]

On Thu, Nov 24, 2005 at 04:42:30PM +0100, Fran?ois-Xavier Coudert wrote:
> > Go with ${PREFIX}/finclude unless pbrook objects.
> 
> I'm not sure this is indeed a good thing in the case where
> PREFIX=/usr. Distros will probably be reluctant to have a new
> subdirectory in /usr that contains only a few files. I'd really rather
> see it go into the ${PREFIX}/lib/... Plus, we have to consider whether
> these files are target-dependent or not. If they are, then they should
> go somewhere labelled with the target triplet (like object files go
> into /usr/lib/gcc/${target-triplet}/${version}/), shouldn't they?

The C preprocessor searches for C++ files:
 /usr/lib/gcc/x86_64-redhat-linux/4.0.1/../../../../include/c++/4.0.1
 /usr/lib/gcc/x86_64-redhat-linux/4.0.1/../../../../include/c++/4.0.1/x86_64-redhat-linux
 /usr/lib/gcc/x86_64-redhat-linux/4.0.1/../../../../include/c++/4.0.1/backward
 /usr/local/include
 /usr/lib/gcc/x86_64-redhat-linux/4.0.1/include
 /usr/include
(and for C just the last 3 dirs).
omp_lib* are written to be ABI neutral, so for multilib configurations
just one set can be installed, but that doesn't have to be the case for
other intrinsic modules or includes (e.g. I guess Fortran 2003 C bindings
will need to be per-ABI thing).
f951 can search multiple directories by default like cpplib does, or,
alternatively, gfortran driver can supply them in lang-specs.h.
The latter could be perhaps easier for the multilib specifc subdir.
We could use a subdir of the same multilib subdir as gcc uses for libgcc.a
etc., e.g.
/usr/lib/gcc/x86_64-redhat-linux/4.1.0/32/finclude
/usr/finclude
for -m32 modules on x86_64.

	Jakub