[gfortran,patch] fix (was: Re: Fortran library question / long double usage / kinds.h)

[FX Coudert]

> Is this something that you think will be done for 4.1?

Here is the first part of the patch to do this. It enables mk-kinds-h.sh 
to create a proper kinds.h header file for libgfortran by matching 
Fortran real kinds and C floating-point types in a general way.

Can someone review it? It's based on the idea that you can determine the 
size of a Fortran kind at compile-time by playing with equivalences, and 
you can do the same for C fp types. Apart from that, it's only scripting 
work. It's regtested on i686-linux, and I tested the script on a few 
non-bash shells (Solaris /bin/sh, ksh, etc.).

OK for 4.1?

And Steve (Ellcey), could you test it? (that is, build libgfortran with 
this patch on ia64-hpux and send me (or the list) the generated kinds.h 
file?

After that, I'll wait for my "large kinds support in library" patch to 
be accepted and go in, and then there will be only small work to have 
complete support for real(10) and real(16) being both present.

By the way (since I'm coming back from holidays, I'm sorry if this mail 
is a bit condensed), Steve (Kargl), did you sort your problems with my 
patch or do you still have a failure? I don't have such a recent 
freebsd, but on a older one it works fine...

-- 
FX