Re-caching GMT remote files #495

seisman · 2020-06-26T23:04:09Z

Description of proposed changes

The earth relief data for GMT 6.0.0 was just changed back to gridline registration. It means that GMT<=6.0.0 uses gridline-registered earth relief data, while GMT>=6.1.0 will uses pixel-registered earth relief data by default.

Related to #451.

Reminders

Run make format and make check to make sure the code follows the style guide.
Add tests for new features or tests that would have caught the bug that you're fixing.
Add new public functions/methods/classes to doc/api/index.rst.
Write detailed docstrings for all functions/methods.
If adding new functionality, add an example to docstrings or tutorials.

weiji14 · 2020-06-26T23:20:14Z

Great, it's just 23 failures instead of 43 now. All of the makecpt ones are passing again. Let's not update the baseline images or numbers just yet, I want to make sure the tests will be forward compatible with #476 and GMT6.1.0 (specifically that we treat the earth_relief grids as Geographic instead of Cartesian).

seisman · 2020-06-26T23:46:37Z

Most of the 23 failures are easy to fix, except two test_grdimage and test_grdimage_slice. The two failures don't make sense.

The following Python script gives me a white-black map with GMT 6.0.0:

import pygmt
from pygmt.datasets import load_earth_relief
pygmt.print_clib_info()

grid = load_earth_relief()
fig = pygmt.Figure()
fig.grdimage(grid, cmap='earth', projection='W0/6i', V='l')
fig.show(method="external")

Output:

Warnings:

grdimage [WARNING]: gmt_grd_project: Output grid extrema [-1.57645e+06/1.0734e+06] exceed extrema of input grid [-8592.5/5559] due to resampling
grdimage [WARNING]: gmt_grd_project: See option -n+c to clip resampled output range to given input range

It even doesn't work with PyGMT v0.1.1 + GMT 6.0.0.

weiji14 · 2020-06-26T23:54:27Z

Yeah, I was getting that a while ago too, sometimes it's purple 😆. I think it's fine in GMT master though.

seisman · 2020-06-27T00:08:47Z

Yes, but when we release PyGMT v0.1.1, everything is fine with GMT 6.0.0.

I checkout PyGMT v0.1.1, and test it with the GMT 6.0.0 I built locally, I still get the white-black map.

seisman · 2020-06-27T00:12:07Z

Just realized that the two tests test_grdimage and test_grdimage_slice pass in the CI jobs, but fail locally for me.

seisman · 2020-06-27T21:22:41Z

FYI, the two tests test_grdimage and test_grdimage_slice suddenly pass locally for me. Perhaps I messed up with my local environment yesterday.

weiji14 · 2020-06-27T22:41:54Z

Hmm, might be that you've cached the pixel registered grids locally, and the cache was just invalidated today. Probably best to run gmt clear cache to be sure.

seisman · 2020-06-27T23:40:46Z

Hmm, might be that you've cached the pixel registered grids locally, and the cache was just invalidated today. Probably best to run gmt clear cache to be sure.

I can't reproduce the failures anymore. They all work well for me.

Perhaps I can update the baseline images and also the numbers to make all the CI green but not merge the branch. It seems easier for us to check GMT master and #476, and decide if we really want a v0.1.2 release.

weiji14 · 2020-06-28T00:27:12Z

Perhaps I can update the baseline images and also the numbers to make all the CI green but not merge the branch. It seems easier for us to check GMT master and #476, and decide if we really want a v0.1.2 release.

Do you mean just cherry-picking the changes here on top of whatever branch we're working on?

seisman · 2020-06-28T00:46:38Z

Here are some tests I just did locally.

Currently, there are 23 failures for GMT 6.0.0. To make all the tests pass, I need to update the following files:

pygmt/clib/session.py
pygmt/tests/baseline/test_grdcontour.png
pygmt/tests/baseline/test_grdcontour_interval_file_full_opts.png
pygmt/tests/baseline/test_grdcontour_labels.png
pygmt/tests/baseline/test_grdcontour_slice.png
pygmt/tests/baseline/test_grdview_on_a_plane.png
pygmt/tests/baseline/test_grdview_on_a_plane_styled_with_facadepen.png
pygmt/tests/baseline/test_grdview_on_a_plane_with_colored_frontal_facade.png
pygmt/tests/baseline/test_grdview_surface_plot_styled_with_contourpen.png
pygmt/tests/baseline/test_grdview_with_cmap_for_perspective_surface_plot.png
pygmt/tests/baseline/test_grdview_with_cmap_for_surface_monochrome_plot.png
pygmt/tests/baseline/test_grdview_with_perspective_and_zaxis_frame.png
pygmt/tests/baseline/test_grdview_with_perspective_and_zscale.png
pygmt/tests/baseline/test_grdview_with_perspective_and_zsize.png
pygmt/tests/test_datasets.py
pygmt/tests/test_grdinfo.py
pygmt/tests/test_grdtrack.py

After these changes, all tests pass. I then temporarily changed @earth_relief_01d to @earth_relief_01d_g (and all other similar dataset) so that I won't see any failures due to grid registration differences. However, I still see 20 failures and have to update the following 15 images:

test_basemap_polar.png
test_coast_iceland.png
test_colorbar_positioned_using_justification_code.png
test_config.png
test_config_font_annot.png
test_config_font_one.png
test_grdview_drapegrid_dataarray.png
test_grdview_grid_file_with_region_subset.png
test_grdview_surface_mesh_plot_styled_with_meshpen.png
test_grdview_surface_plot_styled_with_contourpen.png
test_grdview_with_cmap_for_image_plot.png
test_grdview_with_cmap_for_perspective_surface_plot.png
test_grdview_with_cmap_for_surface_monochrome_plot.png
test_grdview_with_perspective_and_zaxis_frame.png
test_makecpt_to_plot_grid_scaled_with_series.png

Among the 15 failures:

1, 2, 3 are caused by tiny changes in GMT 6.1.0;
4, 5, 6 are surprising to me, possibly related to Let gmt.history and gmt.conf be hierarchical and separate for panels and figures gmt#3523. Need to check if it's a GMT bug before GMT 6.1.0 is released.
7-15 are big differences caused by how CPT hinges are handled in GMT 6.1.0 (Update how CPT hinges are handled gmt#2416). [These grdview images are already updated to make GMT 6.0.0 pass].

Then I updated the 15 images above, cherry-picked the changes in PR #476 (assuming this branch is correct), and ran the tests again. I saw the following failing images:

test_grdview_drapegrid_dataarray.png
test_grdview_grid_dataarray.png
test_grdview_on_a_plane.png
test_grdview_on_a_plane_styled_with_facadepen.png
test_grdview_on_a_plane_with_colored_frontal_facade.png
test_grdview_surface_mesh_plot_styled_with_meshpen.png
test_grdview_surface_plot_styled_with_contourpen.png
test_grdview_with_cmap_for_image_plot.png
test_grdview_with_cmap_for_perspective_surface_plot.png
test_grdview_with_cmap_for_surface_monochrome_plot.png
test_grdview_with_perspective.png
test_grdview_with_perspective_and_zaxis_frame.png
test_grdview_with_perspective_and_zscale.png
test_grdview_with_perspective_and_zsize.png
test_gridimage_over_dateline.png

The 14 grdview tests are failing due to the same reason (unknown yet). Here is an example.
Baseline image

New image if PR #476 is merged:

Perhaps because the default projection depends on the grid type (either GMT_GRID_IS_GEO or GMT_GRID_IS_CARTESIAN.

weiji14 · 2020-06-28T01:06:56Z

Among the 15 failures:

1, 2, 3 are caused by tiny changes in GMT 6.1.0;

4, 5, 6 are surprising to me, possibly related to GenericMappingTools/gmt#3523. Need to check if it's a GMT bug before GMT 6.1.0 is released.

Ok, I think we can agree to not merge in any baseline png image changes, in case anything happens before GMT 6.1.0 release.

7-15 are big differences caused by how CPT hinges are handled in GMT 6.1.0 (GenericMappingTools/gmt#2416). [These grdview images are already updated to make GMT 6.0.0 pass].

Yes, good spotting, we'll definitely need to update these makecpt png images then, unless we want to go for a more future proof solution (i.e. not have the tests depend on earth_relief data). I'm actually quite keen for this bugfix as I've had to manually copy those changed cpt files to make the figures in my last manuscript...

Then I updated the 15 images above, cherry-picked the changes in PR #476 (assuming this branch is correct), and ran the tests again. I saw the following failing images:

The 14 grdview tests are failing due to the same reason (unknown yet). Here is an example.

Perhaps because the default projection depends on the grid type (either GMT_GRID_IS_GEO or GMT_GRID_IS_CARTESIAN.

I wouldn't assume that the branch is correct yet 😆, I still need to make sure things work if node_offset is 0 or missing from the xarray.DataArray. But looking at the two png images, it does like like the projection has changed somewhat. The new image looks to be a bit more curved.

Just to add that it might be possible to keep the same old grdview baseline images if we force those tests to use GMT_GRID_IS_CARTESIAN, at least for the non-coloured ones not using the oleron cpt.

seisman · 2020-06-28T04:08:39Z

Perhaps the best way to minimize image updates is,

Fix the tiny failures about the numbers but don't update any images
Release PyGMT v0.1.2 immediately after GMT 6.1.0 is released, even though some tests still fail due to incorrect baseline images. The PyGMT v0.1.2 should work for both GMT 6.0.0 and 6.1.0 (tests can fail but no crashes).
Bump the minimum required GMT version 6.1.0 and update images.

weiji14 · 2020-06-28T23:24:32Z

Perhaps the best way to minimize image updates is,

Fix the tiny failures about the numbers but don't update any images

I'm okay with this. Could do it in this PR even. I was trying to figure out how to parametrize the tests to use earth_relief_01d on GMT 6.0 and earth_relief_01d_g for GMT 6.1 but it got real messy, and there's too much noise with the failing tests now.

Release PyGMT v0.1.2 immediately after GMT 6.1.0 is released, even though some tests still fail due to incorrect baseline images. The PyGMT v0.1.2 should work for both GMT 6.0.0 and 6.1.0 (tests can fail but no crashes).

No strong opinion on this, I don't really mind when PyGMT v0.1.2 gets released, either before or after GMT 6.1.0 is fine. We should make sure to pin to GMT <= 6.1.0 on conda-forge, i.e. PyGMT v0.1.2 is the last version to support GMT 6.0, and contains some minor changes to make it compatible with GMT 6.1.

3. Bump the minimum required GMT version 6.1.0 and update images.

Looking forward to this 🚀

weiji14 · 2020-07-16T03:15:15Z

I think we can close this since Azure Pipelines isn't used anymore and we've updated to GMT 6.1.0 in #507.

Re-caching GMT remote files

35cc3dd

vercel bot temporarily deployed to Preview June 26, 2020 23:20 Inactive

weiji14 added the maintenance Boring but important stuff for the core devs label Jun 26, 2020

Merge branch 'master' into update-cache

48ad523

vercel bot temporarily deployed to Preview June 27, 2020 23:25 Inactive

weiji14 added the skip-changelog Skip adding Pull Request to changelog label Jun 28, 2020

weiji14 mentioned this pull request Jun 29, 2020

Fix several failures due to updates of earth relief data #498

Merged

5 tasks

weiji14 closed this Jul 16, 2020

seisman deleted the update-cache branch July 16, 2020 03:17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Re-caching GMT remote files #495

Re-caching GMT remote files #495

seisman commented Jun 26, 2020 •

edited

Loading

weiji14 commented Jun 26, 2020 •

edited

Loading

seisman commented Jun 26, 2020

weiji14 commented Jun 26, 2020

seisman commented Jun 27, 2020

seisman commented Jun 27, 2020

seisman commented Jun 27, 2020

weiji14 commented Jun 27, 2020

seisman commented Jun 27, 2020 •

edited

Loading

weiji14 commented Jun 28, 2020

seisman commented Jun 28, 2020

weiji14 commented Jun 28, 2020 •

edited

Loading

seisman commented Jun 28, 2020 •

edited

Loading

weiji14 commented Jun 28, 2020

weiji14 commented Jul 16, 2020

Re-caching GMT remote files #495

Re-caching GMT remote files #495

Conversation

seisman commented Jun 26, 2020 • edited Loading

weiji14 commented Jun 26, 2020 • edited Loading

seisman commented Jun 26, 2020

weiji14 commented Jun 26, 2020

seisman commented Jun 27, 2020

seisman commented Jun 27, 2020

seisman commented Jun 27, 2020

weiji14 commented Jun 27, 2020

seisman commented Jun 27, 2020 • edited Loading

weiji14 commented Jun 28, 2020

seisman commented Jun 28, 2020

weiji14 commented Jun 28, 2020 • edited Loading

seisman commented Jun 28, 2020 • edited Loading

weiji14 commented Jun 28, 2020

weiji14 commented Jul 16, 2020

seisman commented Jun 26, 2020 •

edited

Loading

weiji14 commented Jun 26, 2020 •

edited

Loading

seisman commented Jun 27, 2020 •

edited

Loading

weiji14 commented Jun 28, 2020 •

edited

Loading

seisman commented Jun 28, 2020 •

edited

Loading