fenics team mailing list archive

[Andy Ray Terrel <andy.terrel@xxxxxxxxx>]

Well that got a bit more response than I expected.  Anywho, I'll try
to answer things as I can but I just wanted to point out another model
that I know works well.  To be fair, I realize I'm a little rusty on
FEnCS software dev practices these days and unless my funding
situation radically changes, I don't see me making significant
contributions to the code base.  But I'm happy to show people how to
use it.

On Mon, Oct 31, 2011 at 5:05 AM, Anders Logg <logg@xxxxxxxxx> wrote:
> On Mon, Oct 31, 2011 at 10:24:11AM +0100, Martin Sandve Alnæs wrote:
>> On 31 October 2011 07:26, Johan Hake <johan.hake@xxxxxxxxx> wrote:
>> > On Sunday October 30 2011 22:23:42 Ridgway Scott wrote:
>> >> Maybe we should talk about Andy's comments at the FEniCS meeting
>> >> this week in Lubbock.
>> >
>> > That would be great. But unfortunately not all of us will be there, me
>> > included.
>>
>> Me neither, but the feedback will be welcome, so please send us a summary :)
>>
>> > I would also appreciate a bit more details about the nature and content of
>> > such claims. Obviously it is not very tempting to stand out and raise such
>> > claims directly on a list, but it is difficult to do anything about it if that
>> > is not done.

I've asked a few people to give me more feedback.  I think that there
is a tendency to flame on lists so people generally write off software
that doesn't work for them.  Personally, I like FEniCS because I can
hack my own things together with it but not everyone has written a
thesis with it =P.

>>
>> It would be interesting to know if their problems are basically the
>> same issues we know about and try to address, or something that we do
>> not see from the "inside".
>
> Yes.
>
>> I think we are in general agreement on the main cooperative challenges:
>> - The interface has been changing too much too often, this will
>> hopefully stabilize now, and we will support 1.0 and fix its bugs for
>> a while.
>
> Remember that one of the motivations for the many changes has been to
> get all those changes in place before 1.0 so that we don't need to
> change the interface after 1.0. I'm very happy with the current
> interface and don't foresee any big changes in the near future. We
> also have a quite extensive set of unit tests that check that *all*
> code examples in the Tutorial and DOLFIN chapters in the book will
> continue to run.
>

So I would just say that we want different things.  For example,
rolling your own block solver in FEniCS isn't well streamlined.  Also,
if you want to use a feature in the C++ interface, there is little
recourse to do so in the Python interface.  I believe more in a very
open environment where users help inform such decisions.  Its harder
to keep the software stable, but the end product often suits users
much better.  There are very few software products that come out of
eight years of development without users that are usable by the
masses.  I've seen this with the FLAME group and I think FEniCS is
going to have to work really hard to get a large user base if that is
what it wants.


>> - With the new development plan we try to make the release process less chaotic.
>>
>> As for wanting users, the rapid interface changes may have been a
>> problem, but otherwise I think:
>
> My impression is that we have just the number of users that we want
> and deserve at the moment. It would have been a problem to have 10x
> more users. Now that 1.0 is (almost) here, I think we are ready for
> more users (and developers).
>
>> - Our response time on questions is low, but we do require that the
>> questions make sense.

Yes, this is quite commendable.

>> - We accept patches regularly, but do require that they are in a
>> good state.

Yes, FEniCS is not Trilinos, but nonetheless other people in the
community have the impression that it is difficult to make a patch for
FEniCS.

>
> I think the Launchpad system has good support for contributions. It's
> easy to publish new branches and make merge requests and easy to
> review and merge those branches.
>

I think it has some really nice features.

>> - With the fenics book we have tried hard to make the software well documented.

I think the book is great and regularly tell people to look at it.
But we are really focused on the Python layer and leave the C++ stuff
as an exercise to the reader to translate.

>
> Yes.
>
> --
> Anders
>

At this point, I'm fine with the usual smug "patches welcome" made
famous by SVN, but I'm just parroting things I've heard from the
community.  I'm sure there will be more conversation over some beers
in Lubbock.

-- Andy