MNT: Drop Python 3.4 from test matrix #926

dopplershift · 2017-08-16T16:39:33Z

Conda-forge packages for 3.4 aren't updated any more. Move 3.5 to where
3.4 was used, use 3.6 for "latest everything".

Conda-forge packages for 3.4 aren't updated any more. Move 3.5 to where 3.4 was used, use 3.6 for "latest everything".

dopplershift · 2017-08-16T17:02:00Z

So replacing 3.4 in the build matrix with 3.5 and conda packages means:

NumPy 1.9
Matplotlib 1.4.3
SciPy 0.16.0

Is it ok just to update these?

dopplershift · 2017-08-16T17:03:14Z

And no clue why running on 3.6 breaks 2 tests....

QuLogic · 2017-08-16T21:27:31Z

I personally don't think we should bother with 1.3.1, but I assume this is something that the Met Office has previously had need for.

ajdawson · 2017-08-18T22:01:24Z

1.3.1 seems a little bit too conservative, 1.4.3 is 2 major versions behind the latest and that should satisfy most needs.

These are the latest available for Python 3.5 from conda-forge.

QuLogic · 2017-09-17T23:40:58Z

Any idea what's up with the libgfortran install?

dopplershift · 2017-10-18T23:44:44Z

Need to install libgfortran 1.0 package--which unfortunately isn't an explicit dependence of scipy 0.16 in defaults.

dopplershift · 2017-10-19T00:11:20Z

Well, hey! That last commit at least got Python 3.5 with old packages to pass...

QuLogic · 2017-10-19T22:41:51Z

I think we kind of skipped testing with 1.4.3...

dopplershift · 2017-10-19T22:49:25Z

Down to 1 failure on latest everything for both 2 and 3.

Baseline:

Result for me locally (which also fails with same RMS):

Diff:

Weird.

QuLogic · 2017-10-19T22:51:23Z

I think these are remote images that might just need to be updated.

dopplershift · 2017-10-19T22:51:55Z

I was just looking at that. We could also bump the tolerance for Matplotlib 2.x from 2 to 2.9.

dopplershift · 2017-10-19T23:03:25Z

Holy crap! A green build! Somebody merge it before it breaks again! 😈 🎉

QuLogic · 2017-10-19T23:25:46Z

lib/cartopy/tests/mpl/test_mpl_integration.py

 if MPL_VERSION >= '2':
    _STREAMPLOT_IMAGE = 'streamplot'
-elif MPL_VERSION >= '1.5':
+elif MPL_VERSION >= '1.4.3':
    _STREAMPLOT_IMAGE = 'streamplot_1.5'


Maybe should rename these two?

Done here and in the other file named for 1.5 below.

QuLogic · 2017-10-20T00:05:49Z

.travis.yml

    else
-      wget http://repo.continuum.io/miniconda/Miniconda3-3.7.0-Linux-x86_64.sh -O miniconda.sh;
+      wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh;


Can you make these https too?

QuLogic · 2017-10-20T00:06:42Z

lib/cartopy/tests/test_polygon.py

-        self._assert_bounds(polygon.bounds, 170, 0, 180, 10)
-        polygon = multi_polygon[1]
-        self._assert_bounds(polygon.bounds, -180, 0, -170, 10)
+        poly1, poly2 = multi_polygon


This was sooo annoying.

There's no point in pinning to such an old version of miniconda which uses Python 3.4

Required for at least the old scipy.

Looks like some images were incorrectly matched to matplotlib >=1.5 when really it was 1.4.3 that was the version when things changed. Also update a threshold for a minor difference.

Matplotlib 2.1 now warns about axes() reusing existing instances, so take steps to prevent that.

I'm pretty sure 7 decimal places is sufficient.

These were relying on the order of polygons, which doesn't seem to be guaranteed by the wrapping process. It was essentially relying upon the order of popitem() from a dict, which changed with Python 3.6. This changes to try to identify the appropriate polygons before embarking on testing them.

For now, let's just get the test suite back to green on Travis.

Looks like some of the tiles changed a bit, so bump the tolerance up a bit more so that the test passes.

djkirkham · 2017-11-15T10:01:50Z

I'm getting failures when I run the tests locally: three related to ShapeReader, and a couple of image test failure when running with the latest available Matplotlib (2.1.0). I'll restart the Travis tests here.

Edit: Never mind about the Matplotlib failure, I see that it's been pinned on purpose.

djkirkham · 2017-11-15T10:11:58Z

See https://github.com/djkirkham/cartopy/tree/test_failures for fixes to the ShapeReader tests. Some of them are happening because the records are being received in a different order now and the tests were indexing a list. Another one is happening because the data has apparently been extended and changed slightly (river data).

MNT: Drop Python 3.4 from test matrix

4c014b4

Conda-forge packages for 3.4 aren't updated any more. Move 3.5 to where 3.4 was used, use 3.6 for "latest everything".

dopplershift mentioned this pull request Aug 16, 2017

ENH: Set default zorder for LAND and OCEAN #916

Merged

dopplershift force-pushed the drop-34 branch from 0acea82 to 0bc2076 Compare August 18, 2017 22:21

MNT: Update minimum versions of packages for testing

02c7c60

These are the latest available for Python 3.5 from conda-forge.

dopplershift force-pushed the drop-34 branch from 0bc2076 to 02c7c60 Compare August 18, 2017 22:54

dopplershift force-pushed the drop-34 branch 4 times, most recently from 0ed05c7 to 8f30d3a Compare October 19, 2017 22:40

QuLogic reviewed Oct 20, 2017

View reviewed changes

dopplershift added 8 commits October 20, 2017 11:02

MNT: Don't pin miniconda version

34c00b7

There's no point in pinning to such an old version of miniconda which uses Python 3.4

MNT: Add explicit dependence on libgfortran 1.0

5b952c7

Required for at least the old scipy.

MNT: Update tests to pass on Matplotlib 1.4.3

177e293

Looks like some images were incorrectly matched to matplotlib >=1.5 when really it was 1.4.3 that was the version when things changed. Also update a threshold for a minor difference.

MNT: Fix some warnings from matplotlib

4af389b

Matplotlib 2.1 now warns about axes() reusing existing instances, so take steps to prevent that.

MNT: Silence internal deprecation warning from test

010a8dc

MNT: Fix false failure in test

9c39a9f

I'm pretty sure 7 decimal places is sufficient.

MNT: Pin matplotlib to <2.1 for testing

fe0599d

For now, let's just get the test suite back to green on Travis.

MNT: Bump web tiles test tolerance for matplotlib 2

ff66932

Looks like some of the tiles changed a bit, so bump the tolerance up a bit more so that the test passes.

dopplershift force-pushed the drop-34 branch from cf3104a to ff66932 Compare October 20, 2017 17:02

QuLogic approved these changes Oct 20, 2017

View reviewed changes

has2k1 mentioned this pull request Oct 26, 2017

Fix list parameters passed to artists #910

Closed

dopplershift mentioned this pull request Nov 14, 2017

Catalogue of Test Errors and Failures #952

Closed

pelson added Experience-needed: high SemVer: minor Type: Infrastructure labels Nov 20, 2017

pelson added this to the 0.16 milestone Nov 21, 2017

ajdawson mentioned this pull request Nov 21, 2017

Fix failing tests #957

Closed

djkirkham mentioned this pull request Nov 23, 2017

Fix tests #960

Merged

pp-mo mentioned this pull request Nov 27, 2017

Unpin matplotlib version in mpl v2 testing #961

Closed

pp-mo merged commit ff66932 into SciTools:master Nov 27, 2017

dopplershift deleted the drop-34 branch November 28, 2017 16:38

citibeth mentioned this pull request Oct 24, 2021

Projection Directly from PROJ string / parameters; and fix EPSG bounds #1911

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

MNT: Drop Python 3.4 from test matrix #926

MNT: Drop Python 3.4 from test matrix #926

dopplershift commented Aug 16, 2017

dopplershift commented Aug 16, 2017

dopplershift commented Aug 16, 2017

QuLogic commented Aug 16, 2017

ajdawson commented Aug 18, 2017

QuLogic commented Sep 17, 2017

dopplershift commented Oct 18, 2017

dopplershift commented Oct 19, 2017

QuLogic commented Oct 19, 2017

dopplershift commented Oct 19, 2017

QuLogic commented Oct 19, 2017

dopplershift commented Oct 19, 2017

dopplershift commented Oct 19, 2017

QuLogic Oct 19, 2017

dopplershift Oct 20, 2017

QuLogic Oct 20, 2017

dopplershift Oct 20, 2017

QuLogic Oct 20, 2017

djkirkham commented Nov 15, 2017 •

edited

Loading

djkirkham commented Nov 15, 2017

MNT: Drop Python 3.4 from test matrix #926

MNT: Drop Python 3.4 from test matrix #926

Conversation

dopplershift commented Aug 16, 2017

dopplershift commented Aug 16, 2017

dopplershift commented Aug 16, 2017

QuLogic commented Aug 16, 2017

ajdawson commented Aug 18, 2017

QuLogic commented Sep 17, 2017

dopplershift commented Oct 18, 2017

dopplershift commented Oct 19, 2017

QuLogic commented Oct 19, 2017

dopplershift commented Oct 19, 2017

QuLogic commented Oct 19, 2017

dopplershift commented Oct 19, 2017

dopplershift commented Oct 19, 2017

QuLogic Oct 19, 2017

Choose a reason for hiding this comment

dopplershift Oct 20, 2017

Choose a reason for hiding this comment

QuLogic Oct 20, 2017

Choose a reason for hiding this comment

dopplershift Oct 20, 2017

Choose a reason for hiding this comment

QuLogic Oct 20, 2017

Choose a reason for hiding this comment

djkirkham commented Nov 15, 2017 • edited Loading

djkirkham commented Nov 15, 2017

djkirkham commented Nov 15, 2017 •

edited

Loading