segfault with gfortran > 4.0.3

[John Pye]

Hi again,

Regarding your point about the code's validity. To the best of my
knowledge this code has been tested and compiled with several commercial
Fortran compilers on several different platforms. I also believe it to
be very mature and very well respected code. It has been incorporated
into GAMS, AMPL and other leading mathematical modelling packages. Note
also that the code *does* compile withou problems with gfortran 4.0.3
but crashes gfortran 4.1+. It really looks like a regression to me, not
dodgy code.

Regarding the legal side of this, I can assure you that I have obtained
appropriate permission to post the code in question. It is a very small
subset of his code, just enough to demonstrate the crash. The code I
posted is just support routines (error message output stuff), not his
core algorithms, so it's pretty much worthless by itself. Also, I am
CCing the developer, as you can attest by his details which are online
at http://www-conopt-com. I am sure he will confirm all this, if required.

Perhaps incorporating the code into your test suite may be a problem
(Arne?). But it would really be great if you could have a quick look at
the problem in isolation for now, and perhaps determine what the issue
could be.

I do understand that there are issues here with regard to the GNU
philosophy etc etc, but I also think that I have found a genuine and
serious bug here that needs to be fixed.

Cheers
JP

Steve Kargl wrote:
> On Tue, Aug 01, 2006 at 01:45:24PM +1000, John Pye wrote:
>   
>> There is a test tarball containing just the offending code at the redhat
>> bugzilla, see
>> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=200618
>>
>> I obtained permission from the develop to post that code; please note
>> that, nevertheless, it's not libre code.
>>
>>     
>
> With such a restriction, I will not look at the offending
> code because of fear of becoming tainted.  I don't need 
> some developer knocking on my door claiming I stole his
> code.
>
> Additionally, If I can't produce a testcase to include in the
> gfortran testsuite from the possibly invalid, nonstandard code,
> then it's not worth my time.
>
>   


-- 
John Pye
Department of Mechanical and Manufacturing Engineering
University of New South Wales, Sydney, Australia
http://pye.dyndns.org/