[Bug fortran/52392] New: internal compiler error occurs when compilin files separately but not when all is in one file

["dominique.pelletier at polymtl dot ca" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52392

             Bug #: 52392
           Summary: internal compiler error occurs when compilin files
                    separately but not when all is in one file
    Classification: Unclassified
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dominique.pelletier@polymtl.ca


Created attachment 26758
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26758
contains  the fortran files listed in the above bug report

Greetings,

I have encountered a very strange compiler bug  with gfortran
At least it is to me. I am using gfortran 4.6.1. I used the
comman gffortran --version to get the information.

I am developing a generic finite element program using
object oriented FORTRAN 2003. Initially all sources were
in one file and it worked well: gfortran has no probelms
compiling the programs and its mosules. The results
from executions are corrects. 

I then split the source to have one module per file plus 
a file for the main program. I compiled the individual
modules in the proper sequence but gfortran generated an
internal compiler error message for one of the modules. 
I got the following message from gfortran:

[dp@C4079-13VM1 tests]$ gfortran -c testb_part1.f95
[dp@C4079-13VM1 tests]$ gfortran -c testb_part2.f95
testb_part2.f95: In function âlink_wf_to_zonesâ:
testb_part2.f95:66:0: internal compiler error: in fold_convert_loc, at
fold-const.c:1906
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://bugzilla.redhat.com/bugzilla> for instructions.


I did my best to reproduce the bug on the
simplest example possible. Enclosed are four files
That i compiled on my Lenovo W520 running Linux Fedora 15, inside
VMWare under Windows & enterprise 64 bit version.

testa.f03 : it contains the source that compiles properly
            and executes correctly. I compiled it with

               gfortran testa.f03 -o testa.exe

             and executed it with
               ./testa.exe

testb_part1.f03 and testb_part2.f03 : are obtained by splitting
            testa.f03 . I normaly compile them as follows:

            gfortran - c testb_part1.f03
            gfortran - c testb_part2.f03
            gfortran  testb_part1.o testb_part2.o -o testb.exe

            and got the above internal compiler error message when compiling
            testb_part2.f03

testc.f03 : I merged the two files testb_part1.f03 and testb_part2.f03
            to ensure that i had not introduced any changes when
            I split the original source file into two. I performed
            a diff on testa.f03 and testc.f03 to make sure
            that the two files are identical. I compiled with

               gfortran testc.f03 -o testc.exe

            and executed with
               ./testc.exe

            This yields results that are identical to those obtained
            with testa.f03

The problem seems to  occur when I invoke the pointer-vallued 
function 'get_ptr_to_zone(zone_name), from the module zones_class.
The argument 'zone_name' is a character(len=20)  variable. and
the function 'get_ptr_to zone' is of type 'zone_type'.

Another very similar function 'get_zone_index' is identical
to 'get_ptr_to_zone' except that it returns an index in a vector
of type 'zone_type' instead of a pointer and that it causes
no internal compiler error message.

I thank you in advance for your attention.
Please advise me as how best to proceed with this situation.


Putting all modules in one source file seems to be a viable work-around 
for the short term. I expect that the Fortran 2003 source will grow
quickly enough that keeping all modules in a single source file
will become a hindrance in the near furure. Maybe it is possible to 
extend the viability of this approach  by identifying which modules
or parts of the source must be in the same source file to avoid 
the compiler bug ?

Finally, I would like to know if my limited experience with
fortran 2003, could be the root of my problem? Could I be
trying things that are so twisted that gfortran gets into trouble?

I thoroughly enjoy working with gfortran, and I hope that 
the bug that I am reporting is genuine and not a 'userproblem'.

I look forward to your response .

Sincerely,

Dominique Pelletier
Professor and Canada Research Chair
Mechanical engineering department
Ecole Polytechnique de Montreal

Dominique.Pelletier@polymtl.ca