On my end I think this idea is fine! Although I don't really understand why matplotlib would be needed for that test in the first place. Sorry to have inadvertently introduced a failing test.

That's unrelated, see #5654

Ah, I see. Thanks for merging!

Looks great. I'll merge when tests pass.

Thanks @tomchor. I see this is your first PR here. Welcome to xarray!

Thanks!

I think you spoke too soon though haha. One of the tests is failing and I don't really know why! I never touched the test_chunk() function...:

=========================== short test summary info ============================ FAILED xarray/tests/test_sparse.py::test_chunk - TypeError: operand type(s) a... = 1 failed, 13681 passed, 1110 skipped, 225 xfailed, 67 xpassed, 212 warnings in 445.02s (0:07:25) = **PNC:/usr/share/miniconda/envs/xarray-tests/lib/python3.7/site-packages/PseudoNetCDF/pncwarn.py:24:UserWarning: pyproj could not be found, so IO/API coordinates cannot be converted to lat/lon; to fix, install pyproj or basemap (e.g., `pip install pyproj)` Error: Process completed with exit code 1.

@tomchor thanks. Yes all you need to do is define a test function that checks something and pytest will run it.

you can run it following the instructions here: https://xarray.pydata.org/en/stable/contributing.html#running-the-test-suite

(something like this)

pytest xarray/tests/test_utils.py -k test_latex_name_isnt_split

Thanks! This makes it much easier to test everything locally. I test the new test locally and it should pass now. I also saw that test_chunk() isn't passing and I don't know why. Also I can't run that test locally for some reason.

I defined a function that tests that a long latex sequence stays intact after going through label_from_attrs(), but I'm not sure if that's the same as adding the test. It seems from the Pytest docs that you don't need to call the function explicitly; just define it. In which case that do the trick.

@dcherian could you please confirm?

Also the tests weren't passing before I added this last commit, and I have no idea why.

I wonder if it will look best if we avoid splitting inside the latex code? name, extra and units can have latex as well so I think we need to handle the entire name+extra+units string somehow.

I was playing around a little with regex but I didn't get that far:

```python import textwrap import re

name = r"$Ra_s = \mathrm{mean}(\epsilon_k) / \mu M^2_\infty$" extra = "Longwinded comments about something confusing. " units = r"$[N / m]$" label_raw = name + extra + units label_notex = re.sub("\$(.+?)\$", "X ", label_raw, flags=re.DOTALL) label_notex_wrap = textwrap.wrap(label_notex, 30) label_tex_wrap = ... # regex to add back the latex strings.
```

There might very well be a simpler way of doing this.

I couldn't figure out any way to avoid splitting LaTeX code. I suspect that the way to do it might be kind of complex (but I'm happy if someone can prove me wrong). We can also increase the number of columns for the latex option even more to maintain simplicity...

I'm not entirely sure why that would make the LaTeX renderer fail. But if that's the case and skipping it is an option, I'd test that both the first and last characters are $ before skipping.

Note that for simple latex expressions xarray appears to work fine. For example name = r"$\mathrm{mean}(\epsilon_k)$" works in both figures in the example above.

This definitely sounds okay by me. I'm trying to figure out on my own how to implement this, but these plotting functions are quite complicated with the decorators and the inferences and all! I haven't been able to make much sense of it unfortunately.

@jacobtomlinson Really glad someone's working on this! I'd be glad to help if I can (although I've never contributed to xarray and I don't know much about GPUs). I have some questions though.

Do you have a specific purpose in mind for this? I ask because most other discussions I see related to this really just wanna do ML. However, there's a large user base (myself included) that would benefit immensely from just doing regular (non-machine-learning) operations with a GPU backend.

Also, what's the status on the development? I see no comments after July 2020 and I'm hoping I can help get this back on track if needed!

Wrote a simple function here. I'm not sure what the best to link this to the docs is, since I can't guarantee that this function will change in the future. Maybe link the commit? Or maybe just straight up pasting that into the docs with a small example script using it?

@dcherian That's a good idea. At the moment my implementation is very problem-specific so it's not good to upload that. Generalizing it is very straightforward though (albeit a bit of work), but I'll try to do that soon. I wonder what the best place for that in the docs would be.

Have there been any developments for HDF5 support? I've been trying to read HDF5 data from Dedalus but I've been having a hard time.

I'm not sure where we stand on this issue, but I think since numpy.gradient already exists, it makes more sense to wrap that function for the sake of simplicity instead of making xr.diff differ from the original premise of np.diff.

On the same topic, it bothers me that xr.diff only accepts "upper" and "lower" arguments for the label. The most obvious (and useful) value in my opinion would be "middle", which would correspond to a centered finite differences. Is there any special reason why that option isn't there?

Just realized that you can do that with successive iterations of combine_first. I had misread the docs previously. This can be considered a closed issue.

@maxim-lian But my last point remains: should this really return a zero-valued matrix? I feel like this is counter-intuitive

I just realized this is probably the expected behavior (although counter-intuitive for me) since the coordinates of the points I'm operating on are different. I still don't understand why the returned array has zero values though, instead of nans.

@philippjfr Even with the alteration I still get a HoloMap type not found (using HV 1.9.5) error. I can still see the animation, though. And switching to 'Agg' did allow me to save the animation in mp4, which is good. All workarounds aside, this works pretty well. Thanks a lot to all.

@philippjfr Thanks for the comment. I hand't found anything like that. For me, %%output isn't a recognized magic command, and the renderer.save command fails for me with a tkinter-related error (can't invoke "wm" command), so I really couldn't see anything yet. But I believe I'll be able to solve that eventually.

I didn't know holoviews existed, thanks for pointing that out to me. The fact that it can't create an animation file (as far as I could tell so far) does mean I can't use it, though. So I'll still try to pursue the matplotlib animation option, although maybe for most users holoviews+xarray might already be enough (I don't know).

Unfortunately I don't have any example with DataArray right now. Since I never could take advantage of DataArray's plotting capabilities for animations, I always did animations using pure Numpy. However, I'm talking about the standard matplotlib animations. Here's an example taken from here:

```import numpy as np import matplotlib matplotlib.use("Agg") import matplotlib.pyplot as plt import matplotlib.animation as animation

def update_line(num, data, line): line.set_data(data[..., :num]) return line,

Set up formatting for the movie files

Writer = animation.writers['ffmpeg'] writer = Writer(fps=15, metadata=dict(artist='Me'), bitrate=1800)

fig1 = plt.figure()

data = np.random.rand(2, 25) l, = plt.plot([], [], 'r-') plt.xlim(0, 1) plt.ylim(0, 1) plt.xlabel('x') plt.title('test') line_ani = animation.FuncAnimation(fig1, update_line, 25, fargs=(data, l), interval=50, blit=True) line_ani.save('lines.mp4', writer=writer) ```

If you have some directions on a smart way to bring xarray into the picture, maybe I can try to come up with an example that might evolve into a contribution.