Refine DataTree.equals and DataTree.identical #9473
Conversation
TomNicholas
added
the
topic-DataTree
| other: DataTree, | ||
| nodes_match: Callable[[DataTree, DataTree], bool], | ||
| ) -> bool: | ||
| if not self.isomorphic(other, from_root=True, strict_names=True): |
There was a problem hiding this comment.
I'm wondering if there is a problem with removing the from_root argument but keeping it fixed to True inside check_isomorphic...
Let's imagines I assert_equal two subtrees that have different parents. All the nodes in the subtrees could match, but if the structures above the node where the subtree starts are not isomorphic, it will return False. That doesn't seem like desired behaviour to me - it leaves no way to check that the two subtrees are equal.
| # include inherited coordinates only at the root; otherwise require that | ||
| # everything is also defined at the same level |
There was a problem hiding this comment.
Why do this for identical but not for equals?
There was a problem hiding this comment.
Oh you did talk about this in your comment above
|
|
||
| See Also | ||
| -------- | ||
| Dataset.equals | ||
| DataTree.isomorphic | ||
| DataTree.identical | ||
| """ | ||
| if not self.isomorphic(other, from_root=from_root, strict_names=True): | ||
| to_ds = lambda x: x._to_dataset_view(inherited=True, rebuild_dims=False) |
There was a problem hiding this comment.
Wouldn't it be simpler (and clearer) to use inherited=False? The nodes are checked in depth-first order anyway, so if there is a discrepancy between inherited coordinates it will be discovered at the top of the subtree before checking the children.
I feel like maybe there was some reason why I needed this, in |
xarray/xarray/core/datatree_mapping.py Line 141 in cea354f |
|
Closing this based on the comment in #9215 (comment). Feel free to re-open if you think we do want to keep some part of this PR. |
.identical()now checks for inherited coordinates being defined at the same level ("identical on disk"), whereas.equalsnow checks only that inherited coordinates are equal, even if defined redundantly ("good enough for xarray's alignment").I've also removed the
from_rootargument fromidenticalandequals, though we could bring it back if @TomNicholas feels strongly. It felt like a weird thing to check -- if you want to check from the root, why not just directly compare the root objects instead?We should probably also check that
.pathmatches, in the case of.identicalonly. Currently this is also checked for.equals, which I would propose we remove.