issues

While looking at #4009, I noticed that the combine_attrs kwarg to concat or merge only affects the global attrs of the Dataset, not the attrs of DataArray or Variable members of the Datasets being combined. I think this is a bug.

So far this PR adds tests that reproduce the issue in #4009, and the issue described above. Fixing should be fairly simple: for #4009 pass combine_attrs="drop" to combine_by_coords and combine_nested in open_mfdataset; for this issue insert the combine_attrs handling in an appropriate place - possibly in merge.unique_variable. I'll update with fixes when I get a chance.

• [ ] Closes #4009
• [x] Tests added
• [ ] Passes isort -rc . && black . && mypy . && flake8
• [ ] Fully documented, including whats-new.rst for all changes and api.rst for new API

.plot.surface() method that wraps matplotlib's plot_surface() method from the 3d toolkit. * works with facet grids * disables all the automatic color-map, etc. because it doesn't necessarily make sense for surface plots - the default (in matplotlib, and as implemented now) is not to color the surface. xarray's auto-selection stuff would interfere with plot_surface()'s kwargs.

I'm not sure if there's somewhere good to note the new surface() method in the docs to make it more discoverable? Maybe in one of the examples?

• [x] Closes #5084
• [x] Tests added
• [x] Passes pre-commit run --all-files
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst
• [x] New functions/methods are listed in api.rst

Provides the functionality of scipy.integrate.cumulative_trapezoid.

• [x] Tests added
• [x] Passes pre-commit run --all-files
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst

Use broadcast_like if either x or y inputs are 2d to ensure that both have dimensions in the same order as the DataArray being plotted. Convert to numpy arrays after possibly using broadcast_like. Simplifies code, and fixes #5097 (bug when dimensions have the same size).

@dcherian

IIRC the "resolving intervals" section later will break and is somewhat annoying to fix. This is why the current ugly code exists.

This change seems to 'just work', and unit tests pass. Is there some extra check that needs doing to make sure "resolving intervals" is behaving correctly?

I can't think of a unit test that would have caught #5097, since even when the bug happens, a plot is produced without errors or warnings. If anyone has an idea, suggestions/pushes welcome!

• [x] Closes #5097
• [ ] Tests added
• [x] Passes pre-commit run --all-files
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst

Since @dcherian added Quiver plots in #4407, it's fairly simple to extend the functionality to streamplot().

For example (copying from @dcherian's unit test setup) ``` import xarray as xr from matplotlib import pyplot as plt

das = [ xr.DataArray( np.random.randn(3, 3), dims=["x", "y"], coords=[range(k) for k in [3, 3]], ) for _ in [1, 2] ] ds = xr.Dataset({"u": das[0], "v": das[1]}) ds["mag"] = np.hypot(ds.u, ds.v) ds.plot.streamplot(x="x",y="y",u="u",v="v", hue="mag") plt.show() ```

• [x] Tests added
• [x] Passes pre-commit run --all-files
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst
• [x] New functions/methods are listed in api.rst

Incorrect value "exception" was given in the description of missing_dims argument - "exception" was renamed to "raise".

• [x] Passes isort . && black . && mypy . && flake8

These return dicts of the indices of the minimum or maximum of a DataArray over several dimensions. Inspired by @fujiisoup's work in #1469. With #3871, replaces #1469. Provides a simpler solution to #3160.

Implemented so that da.isel(da.argmin(list_of_dim)) == da.min(list_of_dim) da.isel(da.argmax(list_of_dim)) == da.max(list_of_dim)

• [x] Closes #1469
• [x] Tests added
• [x] Passes isort -rc . && black . && mypy . && flake8
• [x] Fully documented, including whats-new.rst for all changes and api.rst for new API

Don't know if/when I'll have time to finish #4017, so pulling out the fix for #4009 into a separate PR here that is ready to merge.

• [x] Closes #4009
• [x] Tests added
• [x] Passes isort -rc . && black . && mypy . && flake8
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst

combine_attrs argument for merge(), concat(), combine_by_coords() and combine_nested() controls what attributes the result is given. Defaults maintain the current behaviour. Possible values (named following compat arguments) are: - 'drop': empty attrs on returned Dataset. - 'identical': all attrs must be the same on every object. - 'no_conflicts': attrs from all objects are combined, any that have the same name must also have the same value. - 'override': skip comparing and copy attrs from the first dataset to the result.

• [x] Closes #3865
• [x] Tests added
• [x] Passes isort -rc . && black . && mypy . && flake8
• [ ] Fully documented, including whats-new.rst for all changes and api.rst for new API

Note: only added to DataArray.isel() and Variable.isel(). A Dataset should include all dimensions, so presumably it should always be an error to pass a non-existent dimension when slicing a Dataset?

• [x] Closes #3866
• [x] Tests added
• [x] Passes isort -rc . && black . && mypy . && flake8
• [x] Fully documented, including whats-new.rst for all changes and api.rst for new API

xarray does not use OrderedDicts any more, so name did not make sense. As suggested here https://github.com/pydata/xarray/pull/3877#discussion_r396620551.

• [x] Test~~s~~ ~~added~~ updated
• [x] Passes isort -rc . && black . && mypy . && flake8
• [x] Fully documented, including whats-new.rst

Copies the cmap._rgba_bad, cmap._rgba_under, and cmap._rgba_over values to new_cmap, in case they have been set to non-default values. Allows the user to customize plots more by using matplotlib methods on a cmap before passing as an argument to xarray's plotting methods. Previously these settings were overridden by defaults when creating the cmap actually used to make the plot.

I'm not a fan of copying attributes one-by-one like this, but I guess this is an issue with matplotlib's API, unless there's a nicer way to convert a cmap to a discrete cmap than mpl.colors.from_levels_and_colors().

• [x] Closes #3590
• [x] Tests added
• [x] Passes black . && mypy . && flake8