I agree with David and Martin.

Kent


On 8 January 2014 14:38, David Ham <[email protected]> wrote:

> In particular, Firedrake also uses Instant :). Please don't merge instant
> into Dolfin
>
>
> On Wednesday, January 8, 2014, Martin Sandve Alnæs wrote:
>
>>  instant is called by ffc, and also useful (and used) in other
>> non-fenics contexts.
>>
>>  Martin
>>
>>
>> On 8 January 2014 13:07, Garth N. Wells <[email protected]> wrote:
>>
>>> On 2014-01-08 11:33, David Ham wrote:
>>>
>>>> Hi all,
>>>>
>>>> Having discussed this around the Firedrake mob (except Florian who is
>>>> still away), we don't have any objection to UFC going into FFC.
>>>> Indeed, since our FFC branch does indeed have a non-UFC backend, it
>>>> might even make us cleaner and move us towards the point at which we
>>>> can start talking with you about merging our stuff into trunk.
>>>>
>>>> One small issue which will crop up is that FFC uses setuptools while
>>>> UFC has a cmake build process. We would really like a combined package
>>>> to be installable with setuptools (I don't expect this would cause any
>>>> huge issues).
>>>>
>>>>
>>>  I'd suggest that FFC and UFC keep their own config/build systems (with
>>> the C code that crept into FFC being cleaned out), and have a top-level
>>> config/build script for installing both packages and running tests on both
>>> packages.
>>>
>>> With uflacs eventually being merged into FFC, that will leave us with:
>>>
>>> - UFL
>>> - FIAT
>>> - FFC + backends
>>> - Instant
>>> - DOLFIN
>>>
>>> Is Instant used by any projects/packages other than DOLFIN? If not, we
>>> could keep the Instant repo for development and just copy instant/master
>>> into DOLFIN when needed and not bother with Instant releases, leaving us in
>>> terms of user packages and releases with:
>>>
>>> - UFL
>>> - FIAT
>>> - FFC + backends
>>> - DOLFIN
>>>
>>> Four packages is lot better than where we were a year ago (which was 7).
>>>
>>> Garth
>>>
>>>  We would be less happy about UFL going into FFC, as we think that
>>>> breaks an important abstraction. We would be really, really unhappy
>>>> about any of the above being merged with Dolfin, as that would give us
>>>> a Dolfin dependency which is really non-trivial. However neither of
>>>> those merges are being proposed right now, so I'm not sure we need to
>>>> have that discussion now.
>>>>
>>>> Regards,
>>>>
>>>> David
>>>>
>>>> On 8 January 2014 08:03, Martin Sandve Alnæs <[email protected]>
>>>> wrote:
>>>>
>>>>   +1 to merging ufc into ffc.
>>>>>
>>>>> I'd rather not merge in ufl (yet).
>>>>>
>>>>> I plan to merge uflacs into ffc at a later date but not yet.
>>>>>
>>>>> It would be nice if we then split out the compiled stuff from ffc
>>>>> into a separate python module and place all python modules from ufc
>>>>> and ffc in a shared src/ or site-packages/ directory, as this makes
>>>>> it easier to add to python path without installation for running
>>>>> tests.
>>>>>
>>>>> Martin
>>>>> 7. jan. 2014 23:23 skrev "Anders Logg" <[email protected]> følgende:
>>>>>
>>>>>   On Tue, Jan 07, 2014 at 10:12:51PM +0000, Garth N. Wells wrote:
>>>>>>
>>>>>>> We’ve discussed over the past year consolidating FEniCS
>>>>>>>
>>>>>> packages. The motivations are:
>>>>>>
>>>>>>>
>>>>>>> - Fewer packages for users to install
>>>>>>> - Less confusion over dependency versions
>>>>>>> - Simpler development and testing (fewer cross-package
>>>>>>>
>>>>>> dependencies and package tests that depend on other packages)
>>>>>>
>>>>>>> - Reduced burden of making releases (which will hopefully lead
>>>>>>>
>>>>>> to more frequent releases)
>>>>>>
>>>>>>>
>>>>>>> Now that the first FEniCS release from git/Bitbucket has been
>>>>>>>
>>>>>> made,
>>>>>>
>>>>>>> I suggest that we start evolving towards consolidation (rather
>>>>>>>
>>>>>> than
>>>>>>
>>>>>>> taking any radical steps). As a first step, I propose that we
>>>>>>>
>>>>>> merge
>>>>>>
>>>>>>> FFC and UFC into one package. This doesn’t mean that FFC and
>>>>>>>
>>>>>> UFC are
>>>>>>
>>>>>>> suddenly deeply linked, but that UFC becomes one of the
>>>>>>>
>>>>>> implemented
>>>>>>
>>>>>>> FFC targets (and at first, the only).  Longer term, having
>>>>>>> backends/targets in FFC will make the addition of new
>>>>>>>
>>>>>> generation
>>>>>>
>>>>>>> targets easier to develop.
>>>>>>>
>>>>>>> Please respond with thoughts and opinions on merging FFC and
>>>>>>>
>>>>>> UFC!
>>>>>>
>>>>>> I'm very positive to this idea.
>>>>>>
>>>>>> I think UFL could also be merged into the same project. I know
>>>>>> there
>>>>>> will be objections to this from those who only use UFL (David Ham
>>>>>> objected last time I suggested this), but still think it would be
>>>>>> possible to resolve this by adding an option to only install UFL,
>>>>>> something like
>>>>>>
>>>>>> cd ufl && sudo python setup.py install
>>>>>>
>>>>>> Another thing to consider is Debian/Ubuntu packages. I believe
>>>>>> some
>>>>>> work will be involved there as well (to apply for new packages
>>>>>> and
>>>>>> adjust dependencies), so perhaps it would not be optimal to make
>>>>>> many
>>>>>> "small" changes to the package organization? Or is it easy?
>>>>>> Johannes
>>>>>> can comment on this.
>>>>>>
>>>>>> --
>>>>>> Anders
>>>>>> _______________________________________________
>>>>>> fenics mailing list
>>>>>> [email protected]
>>>>>>  http://fenicsproject.org/mailman/listinfo/fenics [1]
>>>>>>
>>>>>
>>>> --
>>>>
>>>> Dr David Ham
>>>> Departments of Mathematics and Computing
>>>> Imperial College London
>>>>
>>>>  http://www.imperial.ac.uk/people/david.ham [2]
>>>>
>>>> Links:
>>>> ------
>>>> [1] http://fenicsproject.org/mailman/listinfo/fenics
>>>> [2] http://www.imperial.ac.uk/people/david.ham
>>>>
>>>>
>>>> _______________________________________________
>>>> fenics mailing list
>>>> [email protected]
>>>> http://fenicsproject.org/mailman/listinfo/fenics
>>>>
>>>
>>
>
> --
> Dr David Ham
> Departments of Mathematics and Computing
> Imperial College London
>
> http://www.imperial.ac.uk/people/david.ham
>
>
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics

Reply via email to