Graphic Tests

[marqh]

I have:

• decreased the check_graphic tolerance to 0.1
• removed 3 tests which are visually reasonable but which now fail with tolerance > 1
• updated two example gallery images, which are failing at tol=0.1
• fixed the iris.util._is_circular function so that it correctly handles -ve diffs using np.abs
  • added unit tests to check _is_circular behaviour
  • updated the CML for the two GRIB load tests, which should be circular, according to the plot images stored (RMS Value: 9.683)
  • these plots now pass with an RMS < 0.1

[DPeterK]

This is a bit odd... It looks like we could either do with some content for these tests, or for them to be removed.

[marqh]

this is an issue of test specialization. this test class is a specialization of a parent class, which implements a range of tests.
Some of these fail, some of these pass.

options:

1. remove all the tests, including passing ones
2. specialize these tests in this class and skip them (implemented)

would an extra developer comment stating that this is a skip pattern for the sub-class be sufficient in this case?

[marqh]

would an extra developer comment stating that this is a skip pattern for the sub-class be sufficient in this case?

(see last commit)

[DPeterK]

@marqh it would be good to get an idea of...

• why the coriolis frequency image has been changed despite the fact that old and new appear identical, and
• why the orca projection 2 image has changed so subtly.

[DPeterK]

I think this is good to go, but I'm just a little reticent to hit merge... @pelson @bjlittle your thoughts also appreciated!

[marqh]

@dkillick I am not sure why such subtle changes in these 2 images occur. I have tried to mimick travis' environment as closely as I can to create these images, perhaps previous contributors had not been so careful, used different versions of cartopy, proj4, etc.

The images are pretty close to the previous ones though

[bjlittle]

@marqh @dkillick Could the subtle graphic shift be related to the change in _is_circular ?

Just a thought ...

[marqh]

@marqh @dkillick Could the subtle graphic shift be related to the change in _is_circular ?

I believe not, one cube is built from scratch, one is loaded from a CF netCDF file, so neither should make use of that utility at all

[QuLogic]

The Coriolis example is a small change in the outline of Seychelles or the Maldives, maybe due to new data. The Orca projection example is a change in scaling, which I think may have originated in Matplotlib somewhere.

[marqh]

obsoleted by #2164 and #2162