Fortran projects (was: Compilation order)

[Ralf Wildenhues]

Hello Alberto,

* Alberto Luaces wrote on Mon, Oct 06, 2008 at 05:11:42PM CEST:
> El Domingo 05 Octubre 2008ES 11:23:40 Ralf Wildenhues escribió:
> > Before you implement manual dependency tracking in your own project:
> > how about implementing it in Automake, so that all Fortran projects can
> > benefit?

> Although I'm a automake newbie, I am interested. I'd be grateful if you gave 
> me those pointers.

> > BTW, is this header dependencies for preprocessed Fortran, or Fortran 90
> > module dependencies?
> 
> This project has only f90 module dependencies, although I also have other 
> older f77 projects that use include files, so I could test both. My idea was 
> to program a Perl script that could find what source files contain which 
> modules and then, in a second pass, write the dependencies of every file 
> given the modules they use. At a later stage, that Perl script could be 
> converted to a shell + sed script if it is needed.

There are two projects for Automake lurking here:

1) support header dependencies for preprocessed Fortran.
This should be a reasonably straightforward extension of side-effect
dependencies currently implemented for C and C++, for example.  The key
ideas are described here:
<http://www.gnu.org/software/automake/manual/html_node/Dependency-Tracking.html>
<http://www.gnu.org/software/automake/manual/html_node/Dependency-Tracking-Evolution.html>

Most that would be needed for ppfc, ppf77 is an adjustment of the
depcomp script for the various ways that can header dependencies
can be output.  And a couple of changes in automake.in to actually
enable this for these languages, and tests to make sure this works.

Norman Gray has done some prior work here IIRC.  Maybe we can just reuse
some of his.


1.5) Well, there is at least one special case issue we should address
separately.  Part of testing (1) would be to check that config.h
inclusion as well as -D settings from DEFS work.

Now, the IBM xlf/xlf90 compiler rejects -DDEFINE but accepts
-WF,-DDEFINE instead.  This can be worked around, either by
some clever substitutions in the makefile, or with a wrapper
script a la `compile', let's call it fppcompile.

(I think I even have a half-baked patch for this lying around somewhere,
the solution strategy seems so familiar.)

1.75) (If there is need for '-c -o' emulation for deficient Fortran
compilers, I definitely have a half-baked patch for that issue.)


2) Support for module dependencies.
Prior external work:
<http://www.theochem.uwa.edu.au/fortran/recompile/>
<http://www.theochem.uwa.edu.au/fortran/make_variables>

Problems: compiling foo.f may generate arbitrary modules bar.mod,
baz.MOD, bla.vo.  So the rules may not be simple.  Or we may want to
limit to some kinds of setups: either some compilers only, or some
required naming strategy.  Anyway, the field is quite open here,
but any kind of support would be cool.


Some general comments: if you choose to contribute to Automake, be our
guest, and welcome!  Please note that nontrivial contributions require
copyright assignment (more off-list if you're interested), and please
develop against the git master version of Automake.  Take a look at the
HACKING file for some more helpful bits.

Cheers,
Ralf