On 14 November 2011 10:26, Marie E. Rognes<meg@xxxxxxxxx> wrote:
On 11/14/2011 11:13 AM, Garth N. Wells wrote:
I frequently break the adaptivity unit test (my fault), but the tests
are not very useful in tracking down problems because they're really
regression tests rather than unit tests. Could they be broken up to
better test individual function calls?
If anyone sees a good way of doing that, yes, definitely! However, the error
estimation features are fairly high-level by construction, so further
(meaningful) simplifications are not entirely obvious.
The test don't have to meaningful - it would be helpful in the case of
a seg fault if we could narrow it down to a particular function call.
It would be a help just to know which function calls are successful
and which fail.
How about moving the adaptivity tests to the regression tests instead, so
that the regression tests include the demos + an additional set of tests?
Would that help?
I think that some low-level tests would be helpful.
