← Back to team overview

dolfin team mailing list archive

Re: Regression in dolfin_utils

 

On 05/26/2011 05:04 PM, Martin Sandve Alnæs wrote:
On 25 May 2011 18:19, Marie E. Rognes<meg@xxxxxxxxx>  wrote:
On 05/25/2011 05:01 PM, Martin Sandve Alnæs wrote:

There has been a regression in the feature set of dolfin_utils, and I
cannot see in the bzr log any sign of this being intentional.


If you are referring to the wrapper code generation (which I assume you
are), yes, it was heavily refactored in the spring. No features were
intentionally removed; however, I did not emphasize keeping features that
didn't seem used.


Form names are ignored, numbers being used instead to name classes
"Form_0" instead of "Form_a" etc.

I find the Form_name naming much more useful than Form_0, and that's
why there is form name input in the first place.


This might very well be a bug rather than an intentional removal.


But the names are still used in the typedef generation, so I'm
guessing ffc doesn't use the form name feature at all?


Probably not. All the generated demo form files for DOLFIN were identical
before and after the rewrite.

Could you add a test for the case(s) you are interested in?

Didn't get around to this today.

The test directories in dolfin are rather sparse,
for example the fem/ unittest folder is empty.
Where am I supposed to place new tests?

All the sfc demos are now functional again in the latest
dev version and some of the cpp ones use this feature.


I simplified your fix (using the form names for naming the forms directly, instead of adding more typedefs). I assume the SyFi buildbot goes red if anything is missing...

--
Marie


References