Subclasses cannot preserve their types during some binary ops and ufuncs combining DataArray and Dataset

Open
#11,829 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@AtharvRG is already working on this.

Since Oct 31, 2026.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
38/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
numpy, python
Domain
backend

Research direction

Read xr.Dataset._binary_op and SupportsArithmetic.__array_ufunc__, then locate the existing tests for binary operations and ufuncs involving DataArray and Dataset. The proposed hooks and their exact API require maintainer agreement, so confirm the intended subclass protocol before editing. Done means the three listed combinations return the specialized subclass type while existing non-subclass behavior remains covered by tests.

Written by the indexing model from the issue text.

Description

needs triage
What is your issue?

From uxarray downstream (UXARRAY/uxarray#1802). While xarray currently discourages subclassing in general, it does also point out uxarray in particular as an example where subclassing is used. In the context of uxarray, it is desirable and intuitive for binary operations and ufuncs involving a uxarray object and an xarray object to return a result which uses the more-specialized uxarray type.

The heart of the issue here is that xarray supports binary operations and binary ufuncs with one input being a Dataset and the other being a DataArray, but currently checks whether it should handle operations directly by checking isinstance(obj, xr.DataArray) or isinstance(obj, xr.Dataset) (which give True even for subclasses). When True, xarray fully handles the operations in a few cases, before subclasses ever get a chance to override anything. These cases are:

  1. xr.Dataset * CustomDataArray, where * can be any binary operation (e.g. multiplication, addition, subtraction)
  2. ufunc(xr.Dataset, CustomDataArray), where ufunc can be any numpy binary ufunc (e.g. np.minimum, np.divmod)
  3. ufunc(xr.DataArray, CustomDataset), where ufunc can be any numpy binary ufunc (e.g. np.minimum, np.divmod)

All other combinations (such as xr.Dataset * CustomDataset, ufunc(xr.DataArray, CustomDataArray), and even xr.DataArray * CustomDataset) can be overridden by the subclass directly, returning the CustomDataset or CustomDataArray type one might hope for (UXARRAY/uxarray#1767 handled this for uxarray). However, these three remaining cases cannot fixed directly in the subclass, because:

  • For (1), the operation goes directly to xr.Dataset's methods, and when xr.Dataset sees that the second term is an instance of xr.DataArray it decides to handle everything, without giving the second term a chance to alter the behavior in any way.
  • For (2) and (3), when numpy applies the ufunc here it sees that the first term is not a subclass of the second term (or vice versa), so it starts checking for any __array_ufunc__ methods on the inputs, in order. This leads to applying the xarray object's __array_ufunc__, which similarly decides to handle everything after seeing that the second term is an instance of an xarray object (xr.DataArray in (2); xr.Dataset in (3)), also without giving the second term a chance to alter the behavior in any way.

One possible solution would be to introduce hooks or flags which do nothing in xarray, but which subclasses could override to regain control. For (1), I believe there would just need to be a hook inside xr.Dataset._binary_op, inside the if isinstance(other, DataArray | Dataset): block. It could probably be something as simple as if getattr(other, "_subclass_overrides_binary_op", False): return NotImplemented. For (2) and (3), I believe there would just need to be a hook inside SupportsArithmetic.__array_ufunc__, something like for x in inputs: if isinstance(x, SupportsArithmetic) and getattr(x, "_subclass_overrides_array_ufunc", False): return NotImplemented.

I would be very interested to open a PR to solve this issue, if xarray maintainers agree that something like this would sound reasonable. Also happy to take any suggestions on this topic, or clarify further if anything isn't clear!

Related: #3980

Dominant language
Python
Stars
4.2k
Forks
1.5k
Avg merge
2d 21h
Merged PRs (30d)
74

Getting set up

Open in Codespaces

Starts the project's dev container in your browser, under your own GitHub account.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from pydata/xarray

All issues in pydata/xarray

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.