thanks @TomNicholas! I'm looking forward to the follow-ups :)

Removed the unnecessary argument.

Okay I've reverted the API, so it basically converts a concat_dim input to it's multidimensional version before continuing.

If we merge this then should I start a separate Pull Request for further discussion about the API?

One of the Travis CI builds failed but again I don't think that was me.

Okay good. So with this API then auto_combine() (and therefore the default behaviour of open_mfdatset()) would be to throw a warning if there isn't enough info in the dataset coords to order them?

Yes, but throwing a warning should probably only be a temporary solution. Long term we should pick a default value (e.g., combine='auto') and require explicitly opting in to different behavior (e.g., combine='manual').

This sounds pretty good to me.

Okay good. So with this API then auto_combine() (and therefore the default behaviour of open_mfdatset()) would be to throw a warning if there isn't enough info in the dataset coords to order them?

OK, do you want to go ahead and revert the documentation/public API for now?

Yes, I'll do that.

I would even be OK supporting nested lists temporarily ... "at your own risk"

If I revert the documentation and you merge this PR then that's exactly what we will have, which would be useful for me until we do the public API.

(Also it seems that whatever was wrong with cftime has been fixed now, as the CI tests are passing.)

The only problem with that idea is that both of these functions should be options for open_mfdataset(), so you still have multiple possible argument structures for that function. I think that's okay though - it would be a lot easier to explain if you say "if the keyword argument combine='auto' then the function auto_combine() is used, if combine='manual' then nested_concat_and_merge() is used, and the filepaths must be structured accordingly."

This sounds pretty good to me.

If you like, we could also merge this work (which is excellent progress towards user-facing APIs) but keep the changes internal to xarray for now until we figure out the public APIs.

That would be great. Then I could start using the master branch of xarray again in my code, while we redo the public API. If I set concat_dims back to concat_dim then it should be completely backwards-compatible.

OK, do you want to go ahead and revert the documentation/public API for now?

I would even be OK supporting nested lists temporarily in xarray via APIs like open_mfdataset and auto_concat without official documentation/support -- using this feature would be "at your own risk". As long as we don't break any existing code, that is totally fine. We just shouldn't advertise an interface until we've happy with it.

I think that would mean there might be some situations that 1D auto_combine() could deal with but nested_concat_and_merge() could not. Situations where you can only merge the variables once they have been individually concatenated. (actually I'm not sure - I would need to think about that)

I think multiple levels nested_concat_and_merge() could always replace auto_combine() as long as merging happens last, at the outer-most level. That's basically what auto_combine() already does.

There are probably some edge cases where the existing auto_combine would you be a little more sloppy about how you put together your lists of variables. But I don't think that's really a good idea for us to support in xarray, since it removes most of the freedom from the implementation.

I personally think a), but I expect users who haven't read the source code for auto_combine() might disagree.

Yes, this is my concern. The way this API handles nested lists makes sense from an implementation perspective, but not really from a user perspective. For users, I think it makes sense to have: 1. The precisely defined nested_concat_and_merge() for cases where you already data sorted. This is pretty common (e.g., with hierarchical directory structures). 2. A magical auto_combine() that inspects coordinates to figure out how to put everything together. We might even rename this to combine() (if we think it's comprehensive enough).

The original version of auto_combine() that I wrote was aiming towards this second use-case, but was obviously fragile and incomplete.

If you like, we could also merge this work (which is excellent progress towards user-facing APIs) but keep the changes internal to xarray for now until we figure out the public APIs.

There are probably some edge cases where the existing auto_combine would you be a little more slopping about how you put together your lists of variables.

I think that's probably the case, but I also think that those edge cases will be so specific that maybe we don't have to explicitly support them. We could just say that anyone who has a combination of datasets that is that funky can just concatenate them themselves?

For users, I think it makes sense to have...

I agree, two separate functions is a lot more intuitive than having auto_combine() behave either auto-magically or totally manually depending on the value of an optional keyword argument.

The only problem with that idea is that both of these functions should be options for open_mfdataset(), so you still have multiple possible argument structures for that function. I think that's okay though - it would be a lot easier to explain if you say "if the keyword argument combine='auto' then the function auto_combine() is used, if combine='manual' then nested_concat_and_merge() is used, and the filepaths must be structured accordingly."

If you like, we could also merge this work (which is excellent progress towards user-facing APIs) but keep the changes internal to xarray for now until we figure out the public APIs.

That would be great. Then I could start using the master branch of xarray again in my code, while we redo the public API. If I set concat_dims back to concat_dim then it should be completely backwards-compatible.

Specifying dim=None will specify (1) explicitly, but otherwise yes, that's exactly what it does.

That is how the current auto_combine behaves too, just for 1D obviously.

Also if you specify dim='something' when that dimension doesn't already exist, you are explicitly telling it to add a new dimension.

I think that would mean there might be some situations that 1D auto_combine() could deal with but nested_concat_and_merge() could not. Situations where you can only merge the variables once they have been individually concatenated. (actually I'm not sure - I would need to think about that)

I think this is a question of whether you think that:

a) auto_combine() is a well-defined operation that is still well-defined in N-D

or

b) auto_combine()'s purpose is to do it's best to combine whatever is thrown at it, and more surgical N-D operations should be the job of another function

I personally think a), but I expect users who haven't read the source code for auto_combine() might disagree.

Do you think it would make sense to try to get rid of "both merging and concatenating" in favor of requiring another dimension in the grid? I guess we do sort of need this for this for open_mfdataset but it feels more complex than desired.

I appreciate that this is pretty complicated, perhaps it should have its own section in the docs (I can do another PR?)

Specifying dim=None will specify (1) explicitly, but otherwise yes, that's exactly what it does.

Yes, this was a typo on my part :).

if the grid of results is auto-combinable along all it's dimensions separately, then the new auto_combine() will auto-magically combine it all

OK, so the rule is that a grid dimension can correspond to any of:

1. merging data variables
2. concatenating along the dimension
3. both merging and concatenating

And to specify case (1) explicitly, you can write dim=None along that dimension?

Thanks for the comments.

What happens if you have a nested list of Dataset objects with different data variables?

This is supported. This new auto_combine() simply applies the old auto_combine() N times along N dimensions, so if the grid of results is auto-combinable along all it's dimensions separately, then the new auto_combine() will auto-magically combine it all, e.g:

```python objs = [[Dataset({'foo': ('x', [0, 1])}), Dataset({'bar': ('x', [10, 20])})], [Dataset({'foo': ('x', [2, 3])}), Dataset({'bar': ('x', [30, 40])})]] expected = Dataset({'foo': ('x', [0, 1, 2, 3]), 'bar': ('x', [10, 20, 30, 40])})

This works

actual = auto_combine(objs, concat_dims=['x', None]) assert_identical(expected, actual)

Also works auto-magically

actual = auto_combine(objs) assert_identical(expected, actual)

Proving it works symmetrically

objs = [[Dataset({'foo': ('x', [0, 1])}), Dataset({'foo': ('x', [2, 3])})], [Dataset({'bar': ('x', [10, 20])}), Dataset({'bar': ('x', [30, 40])})]] actual = auto_combine(objs, concat_dims=[None, 'x']) assert_identical(expected, actual) ``` (I'll add this example as another unit test)

I should point out that there is one way in which this function is not exactly as general as auto_combine() applied N times though. The options compat, data_vars, and coords are specified once for the combining along all dimensions, so you can't currently tell it to use compat='identical' along dim1 and compat='no_conflicts' along dim2, it has to be the same for both. I thought about making these kwargs also accept lists, but although that would be easy to do it seemed like it would complicate the API for a very specific use case?

It might be better to have a separate nested_concat() function rather than to squeeze this all into auto_combine().

That was basically what I tried to do in my first attempt, but nested concatenation without merging along every dimension misses some common use cases, for example if you wanted to auto_combine (or open_mfdataset()) files structured like bash root ├─ time1 │ ├── density.nc │ └── temperature.nc └─ time2 ├── density.nc └── temperature.nc then you would want to merge density.nc and temperature.nc, then concat along 'time'. I suppose you could have nested_auto_combine() as a separate function though. I don't really think that's necessary though, as apart from the substitution concat_dim -> concat_dims then this is fully backwards-compatible.

You might also find it interesting to see how I've used this fork in my own code: I create the grid of datasets here, so that I can combine them here.

I have a question actually - currently if the concat or merge fails, then the error message won't clearly tell you which dimension it was trying to combine along when it failed. Is there a way to do that easily with try... except... statements? Something like python for dim in concat_dims: try: _auto_combine_along_first_dim(...) except MergeError or ValueError as err: raise ValueError(f"Encoutered {err} while trying to combine along dimension {dim}")

(Also something else is breaking in cftime on the python 2.7 builds on AppVeyor now...)

Generally I like the look of this.

What happens if you have a nested list of Dataset objects with different data variables? e.g., a 2D grid of results, with different variables saved in different files. Is this supported at all?

I ask because the current version of auto_combine() was (for better or worse) designed to handle this automagically, but I'm not sure this makes sense for higher dimensional data.

It might be better to have a separate nested_concat() function rather than to squeeze this all into auto_combine().

However, the appveyor builds on python 2.7 are now failing on setup when they try to install cftime, which I don't think was anything I did?

Yes, this is unrelated. See https://github.com/Unidata/cftime/issues/34#issuecomment-443619848; I'm sure it will be fixed soon.

Now that all the tests are passing for all python versions, and I've added an example of multidimensional concatenation to the docstring of auto_combine, I think this is ready to be merged.

However, the appveyor builds on python 2.7 are now failing on setup when they try to install cftime, which I don't think was anything I did?

Okay, I've rewritten it using itertools.groupby instead of toolz.itertoolz.groupby so that there should no longer be a dependence on toolz. The itertools groupby assumes the input is sorted, which makes the code slightly more complex, but I think that's still neater than including the functions and tests required for toolz.itertoolz.groupby.

Itertoolz is a dask.array dependency that we didn't explicitly declare in xarray (yes, this was sloppy). We should probably just copy the function we use into some utils file inside xarray -- it has a compatible open source license. Or we could rewrite to not need it. On Sat, Dec 1, 2018 at 8:18 AM Tom Nicholas notifications@github.com wrote:

I've fixed the code so it works with python 3.5 & 2.7 (tested on my local machine), but the tests on python 2.7 are still failing because for some reason it can't find the itertoolz module. I tried to fix this by adding toolz an explicit dependency in setup.py but that didn't work either.

However, given that xarray is supposed to be dropping python 2 support at the end of this year (), does this particularly matter?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pydata/xarray/pull/2553#issuecomment-443437580, or mute the thread https://github.com/notifications/unsubscribe-auth/ABKS1gVaW7quLcUIo1l-l72xh363f7Utks5u0qvvgaJpZM4YX8wE .

I've fixed the code so it works with python 3.5 & 2.7 (tested on my local machine), but the tests on python 2.7 are still failing because for some reason it can't find the itertoolz module. I tried to fix this by adding toolz an explicit dependency in setup.py but that didn't work either.

However, given that xarray is supposed to be dropping python 2 support at the end of this year (https://github.com/pydata/xarray/issues/1830), does this particularly matter?

This is basically done now - I've implemented everything I wanted to, and included unit tests for the new functionality. I'm also successfully using it in my personal code now.

@shoyer I haven't changed the way the combine function is applied repeatedly to match the implementation in numpy.block(), I think the way I've done it is relatively clear, but let me know if you think it should be refactored.

~~I don't understand why some of the CI tests are failing - all test which run on my machine pass, and the errors in the log seem to come from dask arrays not being loaded, i.e:~~

AssertionError: <xarray.Dataset> Dimensions: (x: 10, y: 8) Dimensions without coordinates: x, y Data variables: foo (x, y) float64 -1.19 0.4476 0.6283 ... -0.09699 -0.2311 -0.6482 <xarray.Dataset> Dimensions: (x: 10, y: 8) Dimensions without coordinates: x, y Data variables: foo (x, y) float64 dask.array<shape=(10, 8), chunksize=(5, 4)>

I should have suggested this before, but the internal implementation of numpy.block() is a nice example of this sort of logic: https://github.com/numpy/numpy/blob/e726c045c72833d5c826e8a13f889746ee0bfdf4/numpy/core/shape_base.py#L661