New pytest test runner#310
Conversation
63c061f to
0d9fb1a
Compare
This also required updating of the regression data as one of the gauges ended up having an extra time point due to unrelated changes. All differences were around 10^-16.
Requires a new argument to test runner build_executable to allow make variables in the command
This is an attempt to support adjoint testing without redoing each adjoint run twice and providing more stability and less fragility. A new helper function has been added to the AMRClaw test python module. Hopefully this will also work in GeoClaw.
0d9fb1a to
c2a9245
Compare
|
Turns out that some of the optimization flags are more sensitive that I had thought. All the tests are passing in the entire matrix now though! Note that for optimization flags that something more aggressive than |
|
Yes, the tests themselves are susceptible to this, but in general this is not an issue. The adjoint tests seem to be the most sensitive, but the acoustics 2D radial problem also seems to have issues. |
|
@mandli: good to know, thanks for tracking this down! |
|
I think I am close to working out all the issues with the new testing framework, but still working on getting test result archiving and a couple of other bugs working. Things that I probably won't get to, but can get added later:
|
|
Merging this along with clawpack/classic#96 and clawpack/geoclaw#697, which has some additional details on tests. |
Uses the new
ClawpackTestRunnerto do testing. See clawpack/clawutil#187 for discussion and details.