fixing behaviour for group parameter in open_datatree

[aladinor]

Hi all.

This might be more complex than pruning the path in the open_group_as_dict function. It is kind of complex because when we use _iter_zarr_groups, it yields the root group. I am still working o it

• Closes open_datatree(group='some_subgroup') returning parent nodes #9665
• Tests added

[aladinor]

@TomNicholas, I think this solution will work, but the DataTree.from_dict(groups_dict) function will create the root as an empty node. I think we might need to check it out. Let me know your thoughts.

[keewis]

I think the reason why you don't get the desired result is that you compute paths relative to the immediate parent of the group, not the global parent. I don't have a deeply nested tree ready for testing (so I can't be sure this actually works), but with the suggestion below I don't get the empty root node anymore.

[aladinor]

I think we are getting close. However, we are still having some discrepancies when comparing both datatrees using the group parameter and when directly selecting it via path. Example:

print(dtree["/group2/subg1"])
Group: /group2/subg1 <----- different root paths here
│   Dimensions:  (x: 2, y: 3)
│   Inherited coordinates:
│     * x        (x) int64 16B -1 -2
│     * y        (y) int64 24B 0 1 2
│   Data variables:
│       blah     (x) int64 16B 2 3
├── Group: /group2/subg1/subsub1 <----- different paths here
│       Dimensions:  (y: 3)
│       Data variables:
│           var      (y) int64 24B 4 5 6
└── Group: /group2/subg1/subsub2 <----- different  paths here

dt2 = xr.open_datatree("test.zarr",
                          group="/group2/subg1"

print(dt2)
Group: /  <----- different root paths here
│   Dimensions:  (x: 2)
│   Dimensions without coordinates: x
│   Data variables:
│       blah     (x) int64 16B ...
├── Group: /subsub1 <----- different  paths here
│       Dimensions:  (y: 3)
│       Dimensions without coordinates: y
│       Data variables:
│           var      (y) int64 24B ...
└── Group: /subsub2 <----- different  paths here

Any comments on this @TomNicholas @keewis?

[keewis]

this is by design, I think? I'd interpret group="/group2/subgroup1" as saying "give me that group as the new root group". Then tree.encoding["source_group"] can contain the full path of the new root.

(if you know unix commands, this would be similar to chroot)

[TomNicholas]

discrepancies when comparing both datatrees using the group parameter and when directly selecting it via path

The reprs are different because the objects are different: dtree["/group2/subg1"] has a parent, whereas dt2 does not. So this is intended.

[TomNicholas]

We just need a test, to remove the encoding bit, and then this is good to go!

[TomNicholas]

The test should look very much like the ones in #9669 - create a tiny nested tree, save it to zarr/netcdf, open it with the group parameter, and check the structure is as expected.

[TomNicholas]

Thank you so much @aladinor !

[keewis]

I've got two comments: one on our strategy on DataTree whats-new entries, and one on the way we compare node datasets.

[TomNicholas]

Sorry @aladinor - in fact could we just do this? #9666 (comment)

[TomNicholas]

This looks good! @keewis 's comments are addressed so I'm going to merge it.

[keewis]

(you can still add yourself to the list of contributors to the DataTree entry in the whats-new, @aladinor Oh, and me too, apparently 😅)

Edit: see

xarray/doc/whats-new.rst

Lines 24 to 32 in 5b2e6f1

	- ``DataTree`` related functionality is now exposed in the main ``xarray`` public
	API. This includes: ``xarray.DataTree``, ``xarray.open_datatree``, ``xarray.open_groups``,
	``xarray.map_over_datasets``, ``xarray.group_subtrees``,
	``xarray.register_datatree_accessor`` and ``xarray.testing.assert_isomorphic``.
	By `Owen Littlejohns <https://github.com/owenlittlejohns>`_,
	`Eni Awowale <https://github.com/eni-awowale>`_,
	`Matt Savoie <https://github.com/flamingbear>`_,
	`Stephan Hoyer <https://github.com/shoyer>`_ and
	`Tom Nicholas <https://github.com/TomNicholas>`_.

[keewis]

Otherwise @TomNicholas can do that while preparing for the release.

[TomNicholas]

Amazing thank you! And thanks for pointing that out so everyone involved can get credit for these great contributions!

[aladinor]

Thanks @TomNicholas and @keewis for your guidance!