Re: F9x and unsupported FC

[Ralf Wildenhues]

Hi Gary,

* Gary Kumfert wrote on Tue, Mar 29, 2005 at 07:06:17PM CEST:
> On Mon, 28 Mar 2005, Ralf Wildenhues wrote:
> > * Norman Gray wrote on Sat, Mar 26, 2005 at 05:51:17PM CET:
> > >
> > > Gary (if you're still on this list): could I get a copy of the FC
> > > support you patched in?
> >
> > me too.  :)
> 
> Its bundled in my research project, Babel, version 0.10.*.
>       http://www.llnl.gov/CASC/components/software.html

It'd be really nice to see this in form of a patch (or several ones)
together with some ChangeLog-like entries describing the changes.
But see below.

> I use it to mix, C, C++, F77, F90, Java, & Python in a single address space
> (no messaging or interpreted middleware). We run on Linux,
> OSX, AIX, and maybe still Solaris.  Make check takes hours!

> > > More generally: if Gary's fixes work OK, can they potentially go into
> > > the libtool distribution (paperwork permitting) without there
> > > necessarily being a formal Fortran Maintainer In Chief?
> >
> > Most certainly.  We take anything that looks like an improvement here
> > and has no obvious drawbacks.
> >
> > I don't think basic support for $FC would be difficult at all --
> > basically just let it do the same thing as for $F77.  I haven't seen
> > any further necessity for changes yet; for example, the Solaris 10
> > Fortran compiler seems to work fine with CVS Autoconf and Libtool,
> > if used as $F77.
> 
> This is roughly what I did.  However, I only did the minimal possible
> to get it to work for me (I was crunching for a release).  You may
> want to review my changes... my understanding of libtool isn't *that*
> deep.

A cursory look showed basically these changes (latter ones AIX-related):
- $FC support which mostly mirrors $F77 support
- one of which might be a bugfix (re aix_use_runtimelinking for tag
  F77 vs CC)
- one looks like a babel-specific hack to enable aix_use_runtimelinking
  always when enable_shared.
- an additional flag runtime_linking_flag abstracting -brtl, and a
  libtool link flag `-dynamic' enabling this.  This change looks very
  bogus and wrong, as -brtl is a configure-time decision in Libtool, and
  also `-dynamic' would be the default mode on most architectures.
  It shows, however, the need for some action here.

Cleaned up, the first two look basically fine for Libtool.  But I would
really, really suggest these patches for branch-2-0 of Libtool or higher
only, because only then is it possible to leave out support for some
tags.  People complain enough that their C-only project suddenly looks
for a C++ compiler (and even fails without one), we don't want another
tag in branch-1-5.

For branch-2-0, the first patch would basically need a more-or-less
rewrite (copying from the current F77 support again).  I will look at
the second one.

If you don't have time to produce individual patches, I can just post
the `diff -u' output.

Thanks for sharing,
Ralf