Do those tests fail with the latest FIAT master, too? --Nico
On Thu, Jul 24, 2014 at 7:27 PM, Aslak Bergersen <[email protected]> wrote: > If you run ffc with this version of fiat then all 1D tests fails with both > python 2 and python 3. Isn't that a problem that > needs to be addressed now? Try running the following (got this from Martin): > > $ more test.ufl > element = FiniteElement("Lagrange", interval, 1) > > ffc --verbose test.ufl > > Aslak > > > 2014-07-24 19:14 GMT+02:00 Nico Schlömer <[email protected]>: > >> Okay, so the issue seems to be that the API for 1D differs from 2D and >> 3D. Consequently, the test needs to look differently, too. >> >> The 1D `tabulate_derivatives()` says: >> """Returns a tuple of length one (A,) such that >> A[i,j] = D phi_i(pts[j]). The tuple is returned for >> compatibility with the interfaces of the triangle and >> tetrahedron expansions.""" >> >> 2D and 3D say: >> # Put data in the required data structure, i.e., >> # k-tuples which contain the value, and the k-1 derivatives >> # (gradient, Hessian, ...) >> >> This should probably be aligned, but the API will break. I would say >> that this needs to be addressed at some point, but the removal of >> ScientificPython/Python3 operability is a different issue. For now, >> you could just adjust the test for 1D. >> >> Cheers, >> Nico >> >> >> On Thu, Jul 24, 2014 at 6:06 PM, Aslak Bergersen >> <[email protected]> wrote: >> > I have added a test for 1D now, you can see it in [2]. >> > >> > Yes, I was talking about [1]. >> > >> > Aslak >> > >> > [2] >> > >> > https://bitbucket.org/aslakbergersen/fiat/branch/aslakbergersen/topic-prepare-py3 >> > >> > >> > 2014-07-24 17:31 GMT+02:00 Nico Schlömer <[email protected]>: >> > >> >> > The code is a little opaque and the returned data structure is a mix >> >> > of >> >> > lists and tuples and >> >> > numpy arrays that differ between 2D and 1D and is not documented >> >> > well. >> >> >> >> Indeed! When I dived into the code it was hard to figure out what data >> >> structure is needed since everything seems quite convoluted at first. >> >> Cleanup and documentation is needed here. >> >> >> >> --Nico >> >> >> >> On Wed, Jul 23, 2014 at 3:28 PM, Martin Sandve Alnæs >> >> <[email protected]> >> >> wrote: >> >> > Note: This is about 1D elements, not linear. >> >> > >> >> > Aslak, can you link to the bitbucket branch where you've fixed some >> >> > of >> >> > the >> >> > other issues with Nicos branch, so others can download it and get to >> >> > the >> >> > issue? >> >> > >> >> > Basically the tabulate_derivative method doesn't return a data >> >> > structure >> >> > in >> >> > the right format so indexing errors occur. The code is a little >> >> > opaque >> >> > and >> >> > the returned data structure is a mix of lists and tuples and numpy >> >> > arrays >> >> > that differ between 2D and 1D and is not documented well. >> >> > >> >> > Martin >> >> > >> >> > >> >> > >> >> > On 23 July 2014 13:59, Aslak Bergersen <[email protected]> >> >> > wrote: >> >> >> >> >> >> Hi! >> >> >> >> >> >> I found an error in your implementation in fiat, Nico. And I'm >> >> >> having >> >> >> some >> >> >> trouble removing it. It is an error for all linear elements (which >> >> >> is >> >> >> not >> >> >> tested by fiat), and can be easy be reconstructed by running >> >> >> >> >> >> element = FiniteElement("Lagrange", interval, 1) >> >> >> >> >> >> The problem seems to be that tabulate_derivative in LineExpansionSet >> >> >> is >> >> >> not changed to return the same as tabulate_derivative in >> >> >> TriangleExpansionSet and TetrahedronExpansionSet. Is there an easy >> >> >> fix >> >> >> for >> >> >> this? >> >> >> >> >> >> Aslak >> >> >> >> >> >> >> >> >> 2014-06-29 22:31 GMT+02:00 Nico Schlömer <[email protected]>: >> >> >> >> >> >>> > Changing idioms >> >> >>> > 2py3 changes idioms that are "outdated". When running the script >> >> >>> > it >> >> >>> > changes >> >> >>> > type(t) != type(q) to not isinstance(t, type(q)). Is this this >> >> >>> > something I >> >> >>> > should do? >> >> >>> > >> >> >>> > Python syntax >> >> >>> > The 2to3 scripts have the possibility to change the comma-syntax >> >> >>> > to >> >> >>> > correct >> >> >>> > python syntax. For example, it changes (a,b) to (a, b). Should I >> >> >>> > run >> >> >>> > this on >> >> >>> > the files as well? >> >> >>> >> >> >>> Those are things that Python2 linters like >> >> >>> >> >> >>> pep8 >> >> >>> pyflakes >> >> >>> flake8 >> >> >>> >> >> >>> usually bring up too. I would say that getting FEniCS clean w.r.t. >> >> >>> to >> >> >>> those three (largely overlapping) improves the code readability and >> >> >>> quality. >> >> >>> >> >> >>> Cheers, >> >> >>> Nico >> >> >>> >> >> >>> >> >> >>> On Fri, Jun 27, 2014 at 9:21 AM, Aslak Bergersen >> >> >>> <[email protected]> wrote: >> >> >>> > Hi! >> >> >>> > >> >> >>> > I have some questions about the supporting to python 3.x. You can >> >> >>> > take >> >> >>> > a >> >> >>> > look at the changes I have done if you want (or need). >> >> >>> > >> >> >>> > Testing with python 3.3 >> >> >>> > I have installed python 3.3 such that I can use it when I want >> >> >>> > (e.g. >> >> >>> > py3 >> >> >>> > script.py). However, when I'm running the tests all the >> >> >>> > dependencies >> >> >>> > are >> >> >>> > missing (For now I'm running python -3). So how do I build it >> >> >>> > with >> >> >>> > python 3? >> >> >>> > >> >> >>> > Support python 3.1 >> >> >>> > callable() returned in python 3.2, so there is no need to change >> >> >>> > it, >> >> >>> > unless >> >> >>> > we want to support python 3.1? >> >> >>> > >> >> >>> > Changing idioms >> >> >>> > 2py3 changes idioms that are "outdated". When running the script >> >> >>> > it >> >> >>> > changes >> >> >>> > type(t) != type(q) to not isinstance(t, type(q)). Is this this >> >> >>> > something I >> >> >>> > should do? >> >> >>> > >> >> >>> > Python syntax >> >> >>> > The 2to3 scripts have the possibility to change the comma-syntax >> >> >>> > to >> >> >>> > correct >> >> >>> > python syntax. For example, it changes (a,b) to (a, b). Should I >> >> >>> > run >> >> >>> > this on >> >> >>> > the files as well? >> >> >>> > >> >> >>> > Six module >> >> >>> > I have used the six modules to make it compatible with 2.x and >> >> >>> > 3.x, >> >> >>> > but >> >> >>> > I'm >> >> >>> > a bit unsure where to put it, or how to properly include it to >> >> >>> > the >> >> >>> > project >> >> >>> > such that all files have access. >> >> >>> > >> >> >>> > -- >> >> >>> > Mvh >> >> >>> > Aslak Bergersen >> >> >>> > 993 22 848 >> >> >>> > >> >> >>> > >> >> >>> > 2014-05-23 12:56 GMT+02:00 Martin Sandve Alnæs >> >> >>> > <[email protected]>: >> >> >>> > >> >> >>> >> UFL doesn't use __metaclass__ but it uses __new__, is the >> >> >>> >> behaviour >> >> >>> >> of >> >> >>> >> that the same? I'd like to clean up those parts at some point >> >> >>> >> but I >> >> >>> >> won't >> >> >>> >> have time before the summer. >> >> >>> >> >> >> >>> >> If we have to change behaviour of Expression we should consider >> >> >>> >> doing >> >> >>> >> that >> >> >>> >> simultaneously with the introduction of an Expression-like ufl >> >> >>> >> type >> >> >>> >> which >> >> >>> >> will have several advantages. >> >> >>> >> >> >> >>> >> Martin >> >> >>> >> >> >> >>> >> >> >> >>> >> On 23 May 2014 12:24, Johan Hake <[email protected]> wrote: >> >> >>> >>> >> >> >>> >>> And then there is the change of syntax for metaclasses in >> >> >>> >>> Python3... >> >> >>> >>> Just >> >> >>> >>> goggle metaclass python 3 and there are several pointers to the >> >> >>> >>> different >> >> >>> >>> syntax. >> >> >>> >>> >> >> >>> >>> Maybe this will be a good point to throw out the usage of >> >> >>> >>> metaclasses >> >> >>> >>> in >> >> >>> >>> DOLFIN? What we need is to add a distinction between >> >> >>> >>> CompiledExpression and >> >> >>> >>> Expression. I have tried this before with no luck ;) >> >> >>> >>> >> >> >>> >>> Johan >> >> >>> >>> >> >> >>> >>> >> >> >>> >>> On Thu, May 22, 2014 at 11:40 AM, Martin Sandve Alnæs >> >> >>> >>> <[email protected]> wrote: >> >> >>> >>>> >> >> >>> >>>> Yes, and if we're lucky we can get to that point without as >> >> >>> >>>> much >> >> >>> >>>> work as >> >> >>> >>>> sympy, since we don't have as much code. >> >> >>> >>>> >> >> >>> >>>> The 2to3 tool can do selective changes like change print "" to >> >> >>> >>>> print("") >> >> >>> >>>> and fix exception syntax, which are compatible with 2.7. >> >> >>> >>>> >> >> >>> >>>> It can also do things like change "a = dict.iteritems()" into >> >> >>> >>>> "a >> >> >>> >>>> = >> >> >>> >>>> dict.items()" which changes the memory usage when run on 2.7. >> >> >>> >>>> These >> >> >>> >>>> differences can instead be resolved by using the python module >> >> >>> >>>> "six" >> >> >>> >>>> which >> >> >>> >>>> implements cross-compatible helper functions for a lot of >> >> >>> >>>> things. >> >> >>> >>>> >> >> >>> >>>> Btw when we switch we should go straight to python 3.3-3.4. >> >> >>> >>>> Supporting 3.0-3.2 side by side with 2.7 is apparently harder. >> >> >>> >>>> >> >> >>> >>>> (Note to Aslak: read the link from Jan!) >> >> >>> >>>> >> >> >>> >>>> Martin >> >> >>> >>>> >> >> >>> >>>> >> >> >>> >>>> On 22 May 2014 11:22, Jan Blechta <[email protected]> >> >> >>> >>>> wrote: >> >> >>> >>>>> >> >> >>> >>>>> Note that there is also an approach of having simultaneously >> >> >>> >>>>> 2.x >> >> >>> >>>>> and >> >> >>> >>>>> 3.x >> >> >>> >>>>> compatible codebase without a need of using 2to3. Allegedly, >> >> >>> >>>>> this >> >> >>> >>>>> is >> >> >>> >>>>> used in SymPy, NumPy and SciPy projects. See >> >> >>> >>>>> >> >> >>> >>>>> >> >> >>> >>>>> >> >> >>> >>>>> >> >> >>> >>>>> http://ondrejcertik.blogspot.cz/2013/08/how-to-support-both-python-2-and-3.html >> >> >>> >>>>> >> >> >>> >>>>> Jan >> >> >>> >>>>> >> >> >>> >>>>> >> >> >>> >>>>> On Thu, 22 May 2014 11:05:43 +0200 >> >> >>> >>>>> Martin Sandve Alnæs <[email protected]> wrote: >> >> >>> >>>>> >> >> >>> >>>>> > The plan for the initial work here is to keep the code >> >> >>> >>>>> > python >> >> >>> >>>>> > 2.7 >> >> >>> >>>>> > compatible but ready for a later swift switch to 3 only. I >> >> >>> >>>>> > suggest we >> >> >>> >>>>> > release fenics 1.5 with python 2.7 compatibility intact but >> >> >>> >>>>> > convertible to python 3 by just running py2to3. Otherwise >> >> >>> >>>>> > there >> >> >>> >>>>> > will >> >> >>> >>>>> > be too much simultaneous breakage. Then we can discuss >> >> >>> >>>>> > whether >> >> >>> >>>>> > we >> >> >>> >>>>> > leave python 2.7 behind in fenics 1.6 or not. >> >> >>> >>>>> > >> >> >>> >>>>> > However, I haven't thought about the swig side in dolfin, >> >> >>> >>>>> > and >> >> >>> >>>>> > as >> >> >>> >>>>> > Johan >> >> >>> >>>>> > mentions keeping the Python CAPI code compatible is not >> >> >>> >>>>> > covered >> >> >>> >>>>> > by >> >> >>> >>>>> > py2to3. I'll discuss this with Johan and Aslak. >> >> >>> >>>>> > >> >> >>> >>>>> > Martin >> >> >>> >>>>> > >> >> >>> >>>>> > >> >> >>> >>>>> > On 22 May 2014 10:49, Garth N. Wells <[email protected]> >> >> >>> >>>>> > wrote: >> >> >>> >>>>> > >> >> >>> >>>>> > > Nice. Do we want to support Python 2.7 and 3, or would it >> >> >>> >>>>> > > be >> >> >>> >>>>> > > more >> >> >>> >>>>> > > sustainable to go all Python 3? My preference is for >> >> >>> >>>>> > > simplicity >> >> >>> >>>>> > > and >> >> >>> >>>>> > > low maintenance, which points to Python 3 only support. >> >> >>> >>>>> > > >> >> >>> >>>>> > > Garth >> >> >>> >>>>> > > On Thu, 22 May, 2014 at 9:39 AM, Martin Sandve Alnæs >> >> >>> >>>>> > > <[email protected]> wrote: >> >> >>> >>>>> > > >> >> >>> >>>>> > >> We have a summer intern at Simula, Aslak Bergersen, >> >> >>> >>>>> > >> who will work on preparations for python 3 support in >> >> >>> >>>>> > >> FEniCS, >> >> >>> >>>>> > >> as well as some other FEniCS tasks, from late June and >> >> >>> >>>>> > >> throughout July. >> >> >>> >>>>> > >> >> >> >>> >>>>> > >> The preparations for python 3 involves mainly: >> >> >>> >>>>> > >> - Replacing ScientificPython for AD in FIAT >> >> >>> >>>>> > >> - Applying and committing backwards compatible parts of >> >> >>> >>>>> > >> py2to3 >> >> >>> >>>>> > >> - Replacing several functions such as dict.iteritems >> >> >>> >>>>> > >> with >> >> >>> >>>>> > >> six.iteritems in UFL and possibly FFC to make sure we >> >> >>> >>>>> > >> keep >> >> >>> >>>>> > >> the >> >> >>> >>>>> > >> same performance and memory behaviour with python 2 and >> >> >>> >>>>> > >> 3. >> >> >>> >>>>> > >> >> >> >>> >>>>> > >> I will be on vacation part of his time here so please >> >> >>> >>>>> > >> help him out if he has questions to the list. >> >> >>> >>>>> > >> >> >> >>> >>>>> > >> Martin >> >> >>> >>>>> > >> >> >> >>> >>>>> > > >> >> >>> >>>>> > > >> >> >>> >>>>> >> >> >>> >>>> >> >> >>> >>>> >> >> >>> >>>> _______________________________________________ >> >> >>> >>>> fenics mailing list >> >> >>> >>>> [email protected] >> >> >>> >>>> http://fenicsproject.org/mailman/listinfo/fenics >> >> >>> >>>> >> >> >>> >>> >> >> >>> >> >> >> >>> > >> >> >>> > >> >> >>> > >> >> >>> > -- >> >> >>> > Mvh >> >> >>> > Aslak Bergersen >> >> >>> > 993 22 848 >> >> >>> > >> >> >>> > >> >> >>> > _______________________________________________ >> >> >>> > fenics mailing list >> >> >>> > [email protected] >> >> >>> > http://fenicsproject.org/mailman/listinfo/fenics >> >> >>> > >> >> >> >> >> >> >> >> >> >> >> >> >> >> >> -- >> >> >> Mvh >> >> >> Aslak Bergersen >> >> >> 993 22 848 >> >> >> >> >> > >> > >> > >> > >> > >> > -- >> > Mvh >> > Aslak Bergersen >> > 993 22 848 >> > > > > > > -- > Mvh > Aslak Bergersen > 993 22 848 > _______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
