ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Jan 22, 2010 at 11:23:48AM +0000, Garth N. Wells wrote:
>
>
> Anders Logg wrote:
> > On Fri, Jan 22, 2010 at 11:42:46AM +0100, Kristian Oelgaard wrote:
> >> Shouldn't we have both 'tensor' and 'quadrature' references like we
> >> had in the old FFC?  Also, before these references can be trusted,
> >> we need to check that the new code will give the same result as
> >> simple_verify_tensors in the old FFC.
> >>
> >> Kristian
> >
> > I've been thinking about this and I think we can actually have just
> > one set of references for the demos. Then the representation will be
> > the one chosen by _auto_select_representation. That will simplify a
> > great deal.
> >
> > But the new regression test will also support having a bunch of forms
> > to test which are not necessarily demos. We can have a file named
> > forms.py which contains a bunch of extra forms to test and in those we
> > can select both representations. This will lead to generation of two
> > different references, like FooTensor.h and FooQuadrature.h and all
> > these references can be thrown into the same directory.
> >
>
> Why bother then with regression tests for the demos?

We need to make sure all the demos are checked so we don't ship a demo
that doesn't work.

> Or just add all the demos to the test forms.

Yes, that's exactly what we are doing. At the start of the regression
tests, all the demos are copied to a temporary directory along with
all the additional tests.

That has two advantages:

1. We make sure we don't forget about adding a demo to the tests when
a demo is added.

2. We can treat demos and additional tests in a uniform way since they
are all just .ufl files after the tests have been generated.

> > This will be simpler than having to iterate over different
> > representations as part of the testing framework. Now we can just
> > iterate over a bunch of form files.
> >
>
> Isn't iterating over different representations easy? Just a change in
> the flag.

A regression testing script quickly becomes very complex (we've had
several before) so any reductions in complexity we can make are
beneficial.

--
Anders


> Garth
>
> > And yes, we need to check against the old simple_verify_tensors. I'll
> > be working on that.
> >

Attachment: signature.asc
Description: Digital signature