What updates done after October 3 may affect gfortran built binaries?

What updates done after October 3 may affect gfortran built binaries?

[Edvardsen Kåre]

This is again related to the failure of execution of a gfortran built
binary ("cannot execute binary", see thread
http://cygwin.com/ml/cygwin/2011-11/msg00034.html )

In short, the main problem is that I can't build a successful binary
from the FLEXPART fortran code (just google "FLEXPART nilu" if you are
curious of what FLEXPART is) on a cygwin installation I did just over a
week ago, but it builds and run without problem on an installation from
October 3. I get no differences in warnings from the bad build compared
to the good one, so I really don't know what to look for.

So far I have come to the conclusion that this must be related to one or
several changes in the cygwin distribution done after October 3. Through
try and failure testing I found that this is not affected by
gfortran/gcc as both gcc 4.3.4 and gcc 4.5.3 works. The latter hangs on
'$EGREP' calls in the 'grib_api' (required library) configure script,
but the workaround of changing to 'egrep' works fine.

I have posted the output from strace, objdump and cygcheck for some of
you to look at in the former thread, but it seem like this is far from a
straight forward problem.

I can see from the [ANNOUNCEMENT] posts that a few things in this cygwin
distro have been updated since October 3 and I kindly ask if someone
have an idea of what updates since then may cause a badly gfortran built
binary if it has nothing to do with gcc alone? 

I will now start going through the updates and change back to versions
yielding October 3 if possible. I think this is important since cygwin
will give the opportunity to run and develop FLEXPART on Windows
machines the way linux-users are used to. In addition, I also see a
potential problem of other fortran software that people want to run
under cygwin.

Regards,
Kåre

Re: What updates done after October 3 may affect gfortran built binaries?

[Marco Atzeri]

On 11/9/2011 1:15 PM, Edvardsen KÃ¥re wrote:
> This is again related to the failure of execution of a gfortran built
> binary ("cannot execute binary", see thread
> http://cygwin.com/ml/cygwin/2011-11/msg00034.html )
>
> In short, the main problem is that I can't build a successful binary
> from the FLEXPART fortran code (just google "FLEXPART nilu" if you are
> curious of what FLEXPART is) on a cygwin installation I did just over a
> week ago, but it builds and run without problem on an installation from
> October 3. I get no differences in warnings from the bad build compared
> to the good one, so I really don't know what to look for.
>
> So far I have come to the conclusion that this must be related to one or
> several changes in the cygwin distribution done after October 3. Through
> try and failure testing I found that this is not affected by
> gfortran/gcc as both gcc 4.3.4 and gcc 4.5.3 works. The latter hangs on
> '$EGREP' calls in the 'grib_api' (required library) configure script,
> but the workaround of changing to 'egrep' works fine.
>
> I have posted the output from strace, objdump and cygcheck for some of
> you to look at in the former thread, but it seem like this is far from a
> straight forward problem.
>
> I can see from the [ANNOUNCEMENT] posts that a few things in this cygwin
> distro have been updated since October 3 and I kindly ask if someone
> have an idea of what updates since then may cause a badly gfortran built
> binary if it has nothing to do with gcc alone?
>
> I will now start going through the updates and change back to versions
> yielding October 3 if possible. I think this is important since cygwin
> will give the opportunity to run and develop FLEXPART on Windows
> machines the way linux-users are used to. In addition, I also see a
> potential problem of other fortran software that people want to run
> under cygwin.
>
> Regards,
> KÃ¥re

my guess binutils
http://cygwin.com/ml/cygwin-announce/2011-10/msg00028.html

or some BLODA.

I downloaded the
http://zardoz.nilu.no/~flexpart/flexpart/flexpart_82-3.tar.gz

but it is not clear to me how to replicate your build.

  make -f makefile.gfs_gfortran_32

fails here:
---------------------------------------------------------------
$ make -f makefile.gfs_gfortran_32
gfortran -O2 -m32 -fconvert=little-endian -frecord-marker=4 
-I/nilu2/home/flexpart/lib/gfortran/include  -c -o writeheader.o 
writeheader.f
includecom:685.22:
     Included at writeheader.f:50:

       common /globalr/             ! REAL
                       1
Error: The equivalence set for 'pplev' cause an invalid extension to 
COMMON 'globalr' at (1)
......
---------------------------------------------------------------

Regards
Marco

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

[SOLVED] Re: What updates done after October 3 may affect gfortran built binaries?

[Dave Korn]

On 09/11/2011 12:15, Edvardsen KÃ¥re wrote:
> This is again related to the failure of execution of a gfortran built
> binary ("cannot execute binary", see thread
> http://cygwin.com/ml/cygwin/2011-11/msg00034.html )
> 
> In short, the main problem is that I can't build a successful binary
> from the FLEXPART fortran code 


  I helped Edvardsen to track this down off-list.  It turns out that FLEXPART
is one of those huge number-crunching Fortran programs that's just jam-packed
with ginormous multi-dimensional arrays.  The final linked executable had 3.38
GB of .bss space!  So, it's not too surprising that it didn't load on 32-bit
Windows; and it's not, as I was worrying, any explicit bug in the compiler or
binutils (although it may be arguable that ld could be helpful if it issued
some kind of warning in these circumstances).

    cheers,
      DaveK

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: [SOLVED] Re: What updates done after October 3 may affect gfortran built binaries?

[Ryan Johnson]

On 15/11/2011 10:53 AM, Dave Korn wrote:
> On 09/11/2011 12:15, Edvardsen KÃ¥re wrote:
>> This is again related to the failure of execution of a gfortran built
>> binary ("cannot execute binary", see thread
>> http://cygwin.com/ml/cygwin/2011-11/msg00034.html )
>>
>> In short, the main problem is that I can't build a successful binary
>> from the FLEXPART fortran code
>
>    I helped Edvardsen to track this down off-list.  It turns out that FLEXPART
> is one of those huge number-crunching Fortran programs that's just jam-packed
> with ginormous multi-dimensional arrays.  The final linked executable had 3.38
> GB of .bss space!  So, it's not too surprising that it didn't load on 32-bit
> Windows; and it's not, as I was worrying, any explicit bug in the compiler or
> binutils (although it may be arguable that ld could be helpful if it issued
> some kind of warning in these circumstances).
Out of curiosity, how then was the OP ever able to make *any* version run?

Ryan


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: [SOLVED] Re: What updates done after October 3 may affect gfortran built binaries?

[Dave Korn]

On 15/11/2011 16:20, Ryan Johnson wrote:
> On 15/11/2011 10:53 AM, Dave Korn wrote:
>> On 09/11/2011 12:15, Edvardsen KÃ¥re wrote:
>>> In short, the main problem is that I can't build a successful binary
>>> from the FLEXPART fortran code

>> FLEXPART is one of those huge number-crunching Fortran programs that's
>> just jam-packed with ginormous multi-dimensional arrays. The final linked
>> executable had 3.38 GB of .bss space!

> Out of curiosity, how then was the OP ever able to make *any* version run?

  Not clear yet but probably owing to changes in one or more of the array
dimensions in the upstream source that he didn't notice taking place.

    cheers,
      DaveK


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: [SOLVED] Re: What updates done after October 3 may affect gfortran built binaries?

[Edvardsen Kåre]

> >> FLEXPART is one of those huge number-crunching Fortran programs that's
> >> just jam-packed with ginormous multi-dimensional arrays. The final linked
> >> executable had 3.38 GB of .bss space!
> 
> > Out of curiosity, how then was the OP ever able to make *any* version run?
> 
>   Not clear yet but probably owing to changes in one or more of the array
> dimensions in the upstream source that he didn't notice taking place.
> 
>     cheers,
>       DaveK

I managed to track down one critical array specification which I sat
smaller before compiling. The 64-bit version use the same initialisation
file as the 32-bit, so an array size of 22E+06 elements instead of 6E+06
was used. The SizeOfUninitializedData is now 5eeba800 (1.6 GB) and
everything works beautiful :)

I would never had thought of this without your help! Thank you very
much!

Cheers,
Kåre

Re: [SOLVED] Re: What updates done after October 3 may affect gfortran built binaries?

[Edvardsen Kåre]

>         >    I helped Edvardsen to track this down off-list.  It turns
>         out that FLEXPART
>         > is one of those huge number-crunching Fortran programs
>         that's just jam-packed
>         > with ginormous multi-dimensional arrays.  The final linked
>         executable had 3.38
>         > GB of .bss space!  So, it's not too surprising that it
>         didn't load on 32-bit
>         > Windows; and it's not, as I was worrying, any explicit bug
>         in the compiler or
>         > binutils (although it may be arguable that ld could be
>         helpful if it issued
>         > some kind of warning in these circumstances).
>         Out of curiosity, how then was the OP ever able to make *any*
>         version run?
>         
>         Ryan

I'm not sure of the exact pre-required settings I had when I compiled
FLEXPART to have a successful executable, but there seem to be various
default parameter settings in some of the FLEXPART include files that
will lead to some ginormous multi dimensional arrays. 
I was pretty sure I did the exact same procedure when I compiled
FLEXPART later and got the non working executable, but if the .bss space
was too large, I must have done something else before. It's just that I
can't possibly think of anything I did different.
However, I'm determined to locate the problem, to advice the FLEXPART
developers. I don't want others to go through something similar.

Kåre