PR17917 gfortran ICE on "equivalence"

[Steve Kargl]

On Mon, Jun 06, 2005 at 09:56:24AM +1000, Bill Northcott wrote:
> I notice this bug got fixed in g95 a couple of weeks ago. g95 will  
> now build a fully working HDF5 with Fortran.

What has been fixed in g95 has no bearings on when or how
things get fixed in gfortran.  Unfortunately, the author
of g95 has chosen not to contribute patches to gfortran
and he has gone to great lengths to ensure porting fixes
from g95 are difficult to integrate into gfortran.

> Is there any chance of a fix for gfortran?

Yes, it will be fixed someday.

> Sorry, I am not up to writing compiler code.

Well, that's the crux of the problem.  I'm not a compiler writer
(just ask Paul Brook :-), but almost every patch that I've submitted
has come from a need.  Yes, it can take several hours of code reading
and several iterations of a patch, but eventually the patch converges
to something acceptable for the source tree.

As to this particular bug, I've briefly looked at it, and it
is nontrivial to fix.  We have to update (if not rewrite) how
*.mod files are written and read, so that the equivalenced variables
are associated.

-- 
Steve