Fixes for Cartopy 0.18 by dopplershift · Pull Request #1372 · Unidata/MetPy

dopplershift · 2020-05-13T05:42:12Z

Description Of Changes

Fix implementation of MetPyMapFeature. We really don't need to be inheriting NaturalEarthFeature, we're pretty much overriding the only thing it does (read and return geometries), so just inherit Feature.
Update tests (for changes to default map boundaries and gridlines)
TODO: Fix clipping issue

Checklist

ClosesCartopy 0.18.0 USCOUNTIES '20m' invalid #1368
Tests added
Fully documented

dopplershift · 2020-05-13T05:44:57Z

I think the fix for MetPyMapFeature here works as a permanent fix and is limited enough for 0.12.2.

CartoPy 0.18 started validating scales for NaturalEarthFeature, which breaks with the scales for our own map features. There's really no reason to inherit from NaturalEarthFeature, so just make this inherit Feature and add a few missing things.

Matplotlib (as of 3.2.2) does not properly clip text to paths. For now, add a clipbox for the Axes bounding box, which will supplement the set clip path.

dopplershift · 2020-05-14T01:08:50Z

Thanks @kgoebber for tracking down the clipping issue. After a couple hours working through things, it turns out that single line is the best way to do this in a future-proof way--until a fix for the matplotlib bug gets in.

0.18 started autoscaling map features and updated some gridline handling, causing changes to our test images.

dopplershift · 2020-07-01T05:47:07Z

Ok, updated a few more image thresholds for tests that had no discernible change in my linux VM. Also fixed some lint as well as a test that was checking for a numpy warning--and failed with 1.19. Let's see how this goes now...

dopplershift · 2020-07-01T07:35:40Z

Now have fixed doctests (pint prints fewer digits) and have also updated for Matplotlib 3.3.0rc1. Let's see if we can get pint to install from git again...because of course that has also broken.

Matplotlib <3.3 checks for closed contours using floating point equality. For *this test*, this was causing different results on macOS vs. Windows/Linux. Fixing this actually changes the labelling of another closed contour as well.

We were using pytest to catch *and check* a warning coming from numpy, which we no longer get. Just use a simple warning filter since it's not part of *our* API anyway.

dopplershift · 2020-07-01T09:23:27Z

Ok, we're close now. A few outstanding failures to address:

test_declarative_barb_gfs and test_declarative_barb_gfs_knots fail on Matplotlib 3.3.0rc1 due to rotations now being done with 64-bit values. No, I'm not kidding, that causes our image differences
test_smooth_window_1d_dataarray and test_gradient_xarray_implicit_axes_transposed fail due to some kind of unit issue with xarray and pint master
A bunch of image tests are failing on matplotlib 2.1; not quite sure why yet or why so many would need threshold adjustments

jthielen · 2020-07-01T14:27:48Z

After doing some git bisect, the two xarray with Pint issues (test_smooth_window_1d_dataarray and test_gradient_xarray_implicit_axes_transposed) appeared with pydata/xarray#3847. In short, xr.testing.assert_allclose will now raise a dimensionality error when trying to compare a DataArray with Quantity inside with one with units on the attribute. This will all likely be adjusted in #1353, so I'm not sure what the best interim fix is for 0.12.2...perhaps testing each important piece of the DataArray ourselves?

This is no longer needed since we're on newer matplotlib versions. Also, now this fixture actually *causes* problems because matplotlib is relying on builtin round returning an int (when not asking for digits), which numpy's does not do.

A few items got deprecated/removed and need updating.

Tripped doc8.

Not sure why this started failing on Travis, but it's only a minor increase. Unable to reproduce locally on a config that should nearly match Travis.

dopplershift · 2020-07-01T21:46:07Z

@jthielen It looks like those tests don't exist on the 0.12.x branch, so I'm guessing they came in with some of your work that's only on master? Assuming that's the case (we'll know for sure when I do a PR to update 0.12.x), we can just wait to fix us with pint/xarray master on #1353. The important part is that we've checked what the problems are.

jthielen · 2020-07-01T22:15:48Z

@dopplershift Yes, now that I look back at it, those were added in #1223 and #1260 respectively, so I'll just plan on addressing it in #1353.

jthielen · 2020-07-09T16:54:30Z

In coming back to this, I'll plan on fixing those xarray tests on master in #1378 instead, since that PR comes before #1353. Though, in digging through things now, hgrecco/pint#1123 (recently merged in Pint) is breaking a fair number of our EL/LFC/CAPE/CIN tests in thermo. Where would you want to include a patch for that?

dopplershift · 2020-07-09T19:02:39Z

Since it's only Pint master (right?), we should be able to get this green without it. All that's waiting on is me to figure out what's going on with 2.1 (I'm just going to update the images for the matplotlib 3.3 RC).

What's weird about matplotlib 2.1 is this used to pass and I don't think this changed, so I was trying to investigate whether they were falsely passing before.

jthielen · 2020-07-09T19:05:01Z

Since it's only Pint master (right?), we should be able to get this green without it.

Correct. So would a separate PR to master (which you can then cherry pick commits to 0.12.x) be good?

jthielen · 2020-07-27T15:31:27Z

@dopplershift Were all the fixes here incorporated in #1416, and if so, can this be closed?

dopplershift · 2020-07-27T16:31:49Z

All of the Cartopy-0.18-relevant fixes were in #1416. There are of course several other things that will go in other PRs, but I don't want to close this until they're all in.

dopplershift · 2020-08-05T05:33:35Z

All of this went in in pieces, among them #1411, #1413, #1416, #1420, and #1422. (I checked by rebasing on current master.)

dopplershift requested a review from dcamron as a code owner May 13, 2020 05:42

dopplershift added this to the 0.12.2 milestone May 13, 2020

dopplershift added 2 commits May 13, 2020 19:07

ENH: Fix MetPyMapFeature for CartoPy 0.18
ab26406
CartoPy 0.18 started validating scales for NaturalEarthFeature, which breaks with the scales for our own map features. There's really no reason to inherit from NaturalEarthFeature, so just make this inherit Feature and add a few missing things.

MNT: Work-around matplotlib text clipping issue
5bc34f3
Matplotlib (as of 3.2.2) does not properly clip text to paths. For now, add a clipbox for the Axes bounding box, which will supplement the set clip path.

dopplershift force-pushed the cartopy-0.18 branch from ba27d89 to 9999d76Compare May 14, 2020 01:07

This was referenced May 25, 2020
Xarray accessor updates to remove automatic quantification and in-place operations #1378
Closed
Improved xarray compatibility on function input and output #1353
Closed
Quick patch to bypass scale validation in new Cartopy #1371
Closed

dopplershift force-pushed the cartopy-0.18 branch 3 times, most recently from 2822b84 to 2625e99Compare July 1, 2020 05:09

MNT: Fix tests for CartoPy 0.18
44aa6ce
0.18 started autoscaling map features and updated some gridline handling, causing changes to our test images.

dopplershift force-pushed the cartopy-0.18 branch from 2625e99 to 2fe93c2Compare July 1, 2020 05:45

dopplershift force-pushed the cartopy-0.18 branch from 2fe93c2 to cf9e003Compare July 1, 2020 07:33

dopplershift added 3 commits July 1, 2020 02:53

MNT: Fix contour test consistency
7659ceb
Matplotlib <3.3 checks for closed contours using floating point equality. For *this test*, this was causing different results on macOS vs. Windows/Linux. Fixing this actually changes the labelling of another closed contour as well.

MNT: Fix up lint found by latest plugins
0cc1d91

MNT: Fix test for NumPy 1.19
65565e7
We were using pytest to catch *and check* a warning coming from numpy, which we no longer get. Just use a simple warning filter since it's not part of *our* API anyway.

dopplershift force-pushed the cartopy-0.18 branch 2 times, most recently from 894f15f to 12c7582Compare July 1, 2020 09:22

dopplershift force-pushed the cartopy-0.18 branch from 12c7582 to 508948aCompare July 1, 2020 19:44

dopplershift added 4 commits July 1, 2020 15:39

MNT: Update redirecting link
8a8c6af

MNT: Fix doctests for updates to Pint's printing
ec22b94

MNT: Remove uses of patch_round fixture
46f94b5
This is no longer needed since we're on newer matplotlib versions. Also, now this fixture actually *causes* problems because matplotlib is relying on builtin round returning an int (when not asking for digits), which numpy's does not do.

MNT: Updates for matplotlib 3.3
4f8496b
A few items got deprecated/removed and need updating.

dopplershift added 3 commits July 1, 2020 15:39

MNT: Fix installing pint from git on Travis
f19f721

MNT: Wrap some long lines in references
edf5ac7
Tripped doc8.

MNT: Tweak image tolerance
5c00528
Not sure why this started failing on Travis, but it's only a minor increase. Unable to reproduce locally on a config that should nearly match Travis.

dopplershift force-pushed the cartopy-0.18 branch from 508948a to 5c00528Compare July 1, 2020 21:39

jthielen mentioned this pull request Jul 9, 2020
Fix _check_argument_units for pint __eq__ returning duck array #1398
Closed

jthielen mentioned this pull request Jul 27, 2020
MNT:(deps): Bump xarray from 0.15.1 to 0.16.0 in /ci #1419
Closed

dopplershift marked this pull request as draft July 27, 2020 16:32

dopplershift closed this Aug 5, 2020

dopplershift deleted the cartopy-0.18 branch August 5, 2020 05:33

dopplershift removed this from the 0.12.2 milestone Aug 5, 2020

Uh oh!

Conversation

dopplershift commented May 13, 2020

Description Of Changes

Checklist

Uh oh!

dopplershift commented May 13, 2020

Uh oh!

dopplershift commented May 14, 2020

Uh oh!

dopplershift commented Jul 1, 2020

Uh oh!

dopplershift commented Jul 1, 2020

Uh oh!

dopplershift commented Jul 1, 2020

Uh oh!

jthielen commented Jul 1, 2020

Uh oh!

dopplershift commented Jul 1, 2020

Uh oh!

jthielen commented Jul 1, 2020

Uh oh!

jthielen commented Jul 9, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dopplershift commented Jul 9, 2020

Uh oh!

jthielen commented Jul 9, 2020

Uh oh!

jthielen commented Jul 27, 2020

Uh oh!

dopplershift commented Jul 27, 2020

Uh oh!

dopplershift commented Aug 5, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jthielen commented Jul 9, 2020•
edited
Loading