dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Johannes Ring wrote:
> On Tue, Feb 16, 2010 at 4:44 PM, Garth N. Wells <gnw20@xxxxxxxxx> wrote:
>>
>> Anders Logg wrote:
>>> On Tue, Feb 16, 2010 at 03:38:18PM +0000, Garth N. Wells wrote:
>>>> Anders Logg wrote:
>>>>> On Tue, Feb 16, 2010 at 04:29:24PM +0100, Anders Logg wrote:
>>>>>> On Tue, Feb 16, 2010 at 01:23:33PM +0100, Johannes Ring wrote:
>>>>>>> On Tue, Feb 16, 2010 at 12:49 PM, Harish Narayanan
>>>>>>> <harish.mlists@xxxxxxxxx> wrote:
>>>>>>>> On 2/16/10 10:27 AM, Johannes Ring wrote:
>>>>>>>>> On Tue, Feb 16, 2010 at 10:37 AM, Harish Narayanan
>>>>>>>>> <harish.mlists@xxxxxxxxx> wrote:
>>>>>>>>>> On 2/16/10 9:06 AM, Garth N. Wells wrote:
>>>>>>>>>>> Anders Logg wrote:
>>>>>>>>>>>> On Tue, Feb 16, 2010 at 08:42:55AM +0000, Garth N. Wells wrote:
>>>>>>>>>>>>> Anders Logg wrote:
>>>>>>>>>>>>>> On Tue, Feb 16, 2010 at 09:09:31AM +0100, Johannes Ring wrote:
>>>>>>>>>>>>>>> On Mon, Feb 15, 2010 at 6:06 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>>>>>>>>>>>>>>> On Mon, Feb 15, 2010 at 03:49:05PM +0100, Anders Logg wrote:
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>   UFL    0.5.2
>>>>>>>>>>>>>>>>>>>>>>>   FErari 0.2.0
>>>>>>>>>>>>>>>>>>>>>>>   FFC    0.9.1
>>>>>>>>>>>>>>>> Now released.
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>   Instant 0.9.8: Why is the buildbot failing?
>>>>>>>>>>>>>>>> Now released.
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>   DOLFIN  0.9.7: SCOTCH problems
>>>>>>>>>>>>>>>> Remains to fix:
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>  - Drop Hardy support (upgrade buildbots)
>>>>>>>>>>>>>>> This might take some time, at least for linux64-exp since I have no
>>>>>>>>>>>>>>> control over this. I guess it will be upgraded when Lucid is out.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> What should we install on hardy-i386? Karmic or perhaps Lucid?
>>>>>>>>>>>>>> I think we should choose the simplest option. I don't know what is
>>>>>>>>>>>>>> easiest and fastest, either dropping Hardy support (which requires
>>>>>>>>>>>>>> upgrades of buildbots and some extra administration) or adding the
>>>>>>>>>>>>>> required #ifdefs for unordered_set/set. Garth?
>>>>>>>>>>>>>>
>>>>>>>>>>>>> I don't want to add ifdefs (it actually involves more than just that).
>>>>>>>>>>>>> Harish told me yesterday that he wasn't using the standard OSX gcc (his
>>>>>>>>>>>>> old version didn't support tr1 well), which means we could probably
>>>>>>>>>>>>> switch back to using the tr1 unordered containers and not break Hardy or
>>>>>>>>>>>>> standard standard OSX installations.
>>>>>>>>>>>>>
>>>>>>>>>>>>> I would still suggest that we drop Hardy in the near future, perhaps
>>>>>>>>>>>>> once Lucid is out.
>>>>>>>>>>>> That seems like a good plan since Lucid is the next LTS release.
>>>>>>>>>>>>
>>>>>>>>>>>> Do you know which #ifdefs to add so we can get the buildbot green?
>>>>>>>>>>>>
>>>>>>>>>>> No ifdefs. Just change
>>>>>>>>>>>
>>>>>>>>>>>   boost::unordered_set -> std::tr1::unordered_set
>>>>>>>>>>>   boost::unordered_map -> std::tr1::unordered_map
>>>>>>>>>>>
>>>>>>>>>>> and the relevant includes.
>>>>>>>>>> This should work, except it might break the mac buildbot. While I wasn't
>>>>>>>>>> using the most recent OS X or gcc, I think I was in line with the mac
>>>>>>>>>> buildbot.
>>>>>>>>>>
>>>>>>>>>> Johannes, what does the buildbot run? Leopard or Snow Leopard, and what
>>>>>>>>>> gcc does it use?
>>>>>>>>> I think it was Leopard and gcc 4.3 from fink. However, the mac
>>>>>>>>> buildbot has been offline for some weeks now. We have bought a new
>>>>>>>>> imac to be used as a dedicated mac buildbot but I haven't started to
>>>>>>>>> install anything on it yet. I'm not sure if we should go for fink or
>>>>>>>>> macports. Any recommendations?
>>>>>>>> I suggest MacPorts. It is a bit more unstable than fink, but it allows
>>>>>>>> for greater control and has a very recent collection of packages.
>>>>>>> Thanks, I will look at MacPorts when I setup the new buildbot.
>>>>>>>
>>>>>>> Johannes
>>>>>> So is the conclusion that std::tr1::unordered_set will work with
>>>>>>
>>>>>>   1. Modern operating systems
>>>>>>   2. Ubuntu Hardy
>>>>>>   3. New Mac versions (which is what we will run on the buildbot)
>>>>>>
>>>>>> ?
>>>>>>
>>>>>> If so I (or someone else) can switch to std::tr1::unordered_set.
>>>>> And the scotch/parmetis include thing needs to be fixed. Garth?
>>>>>
>>>> I can take a look tonight, although I haven't followed closely what the
>>>> problem is.
>>> Great. There seems to be a conflict between two versions of
>>> parmetis.h, the ParMETIS one and another installed by SCOTCH.
>>>
>>> I think the solution is to add a scotch/ prefix in the #include
>>> and modify the include path accordingly so it doesn't pick up
>>> parmetis.h from SCOTCH when the real one is needed.
>>>
>> How can I reproduce the problem? I don't see it on my machine. Sounds
>> strange that SCOTCH includes parmetis.h. Is it the compatibility
>> functions that SCOTCH provides that cause the problem?
> 
> This is how I build SCOTCH and PT-SCOTCH on the buildbot:
> 
>   wget http://gforge.inria.fr/frs/download.php/23390/scotch_5.1.7.tar.gz
>   tar zxf scotch_5.1.7.tar.gz
>   cd scotch_5.1/src
>   ln -s Make.inc/Makefile.inc.i686_pc_linux2 Makefile.inc
>   sed -i "/^CCD/s/gcc/mpicc/g" Makefile.in
>   sed -i "/^CFLAGS/s/= /= -fPIC /g" Makefile.inc
>   make
>   make ptscotch
>   cd ..
>   export SCOTCH_DIR=$PWD
> 
> Then I have
> 
>   ls $SCOTCH_DIR/include
>   metis.h  parmetis.h  ptscotchf.h  ptscotch.h  scotchf.h  scotch.h
> 
> But it seems that if I do "make install" it won't copy metis.h and
> parmetis.h to $prefix/include so I guess that is the solution.
> 

That's right. Hopefully this fixes the problem.

Garth

> Johannes