[mpich2-maint] #261: mpich2, 1.0.8 release: a crash with Fortran90?

[jdelia@intec.unl.edu.ar]

Hi Tobias,

Tobias Burnus <burnus@net-b.de> ha escrito:

> Hi,
>
> jdelia@intec.unl.edu.ar wrote:
>> Thanks for your suggestion. As a summary of the mpich2 building of the
>> current 1.0.8 release, we have:
>> 1) with the gfortran version of the system: it's crash.
>> GNU Fortran 95 (GCC) 4.1.0 20060515 (Red Hat 4.1.0-18)
>>
>> 2) with the current (beta) gortran release:  it's OK.
>> GNU Fortran (GCC) 4.4.0 20081021 (experimental) [trunk revision 141258]
>
> I'm not sure how to read that email (from a gfortran perspective). Is
> this supposed to be a gfortran bug report?

No, I don't think so. It would to be seen as informative only for  
other interested users.

> From the GCC perspective only the following compilers are supported:
> - Current release series, 4.3.2
> - Previous release series, 4.2.4
> - active development, 4.4.0
>
> 4.1.x was the first usable version of gfortran but 4.2.x and later
> contain several bug fixes; especially the 4.1-Zero-Version was not in
> the best shape. (Remember: gfortran was as new compiler added only with
> 4.0.0; unfortunately, it takes some time to iron out bugs.)
> Additionally, note that bugs in older versions than the
> active-development version (currently 4.4) will only be fixed if either
> (a) it is a regression or (b) a serious bug (e.g. esp. wrong bugs) - but
> only if the fix is simple. The reason is that one easily can introduce
> regressions, which should never ever happen in the release branches.

OK. I suspected this bug-fixing policy.

> I would suggest that you update to a newer compiler than 4.1.0 as you
> seemingly did.

Yes. We plain to do a system upgrade in this cluster.

> Some remarks:
> - gzipping a TAR archive, which contains only gzipped files does not
> make much sense; either only "tar" or tar-zip but then don't zip the
> contained files.

I am sorry. I relaxed with this redundancy.

> - When reporting a bug, one needs more information, e.g. where to get
> the source code, how it fails and how to reproduce it, etc. (A small
> test case would also be handy.)

Yes, I know this...

Regards,
Jorge.
--

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.