[MITgcm-devel] testreport and cross compiling

[Martin Losch]

Hi Ed,

don't stay up until 3am again and worry about this. The problem is  
somewhere with the OS (I believe). Tomorrow (if I will have  
recuperated from my cold), I'll set up a test for this and really  
look at it in detail.

The strange thing is that genmake2 *DOES NOT ALWAYS* get the correct  
value of CC from the source build-options file. The emphasis is on  
"not always". The problem is not reproducable and very strange. And  
of course, testreport will not really work for me, if I don't sort  
out this issue first. (as usual my sysadmins are not very helpful  
here, as they don't see a sense in these tests, ie, don't see why  
they should spend their time on fixing something that can be avoided  
with a workaround (see the link thing)).

Talk to you later,

Martin
On 10 Feb 2008, at 18:12, Ed Hill wrote:

> On Sun, 10 Feb 2008 16:44:41 +0100 Martin Losch <Martin.Losch at awi.de>
> wrote
>
>> Hi Jean-Michel,
>>
>> thanks for making that clear. I had hoped that it worked that way.
>> The sad circumstance is that when I have CC set to some cc or gcc,
>> then mpack and the comparison code compile (with cc or gcc), but the
>> stupid machine does not pick up the value of CC from the build
>> options file for genmake2. The result is that not only do all sorts
>> of tests fail, the compilation even hangs on one of the test. What's
>> even more disturbing, is that this does not happen all the time, but
>> only in some directories of the file system. I have not yet detected
>> a pattern. When I run genmake2 "by hand" I do "ln -s `which sxcc`
>> cc" first so that genmake2 used cc but really sxcc (the cross
>> compiler), a subsequent make CLEAN removes all links (also the one to
>> sxcc), and a "make makefile" works fine. But this does not work with
>> testreport, as you can imagine. So I tried setting CC=sxcc in
>> my .login/.cshrc files and that works, but then testreport uses that
>> CC as well. I guess I have to trace down, why genmake2 does not see
>> the CC value specified in the build options file.
>
>
> Hi Martin,
>
> I'm confused by the above description.  How, exactly, are you  
> setting CC
> for testreport?  If you use an environment variable for CC [note that
> this is {ba,}sh syntax]:
>
>   export CC=gcc
>   ./testreport
>
> then you should have CC defined with one value for testreport and
> genmake2 will then (as it always does) get a potentially different CC
> value by source-ing the specified optfile.  So, the result is:
>
>   for testreport:  CC=gcc (or whatever your IA64 machine prefers)
>   for genmake2:    CC=...whatever was in the specified optfile...
>
> Does that make sense?
>
> If you are *NOT* seeing the above then something is going wrong  
> because
> I've looked at the code and testreport clearly gets CC as an  
> environment
> variable (with a default value of "cc" if nothing is provided) and
> genmake2 clearly gets its variables by source-ing the specified
> optfile.  So if you are not getting the above then please send the
> *EXACT* commands you are using.  And please include a copy of all your
> environment variables ["env > ML_env_vars"] written to a file
> immediately before you run testreport.
>
>
>> For Ed: The headnode/front end is a TX7 with uname -a:
>> Linux sx8 2.6.5-7.282-default #1 SMP Tue Aug 29 10:40:40 UTC 2006
>> ia64 ia64 ia64 GNU/Linux
>
> OK, so you have an SX machine with a Linux-IA64 "head" node.  You  
> should
> have basically no trouble building testreport's mpack and the  
> comparison
> code to run on the IA64 "head" node while everything else is built to
> run on the SX.
>
> Ed
>
> -- 
> Edward H. Hill III, PhD  |  ed at eh3.com  |  http://eh3.com/
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel