Enable mixins behavior #1437
Enable mixins behavior #1437
Conversation
Codecov Report
|
for more information, see https://pre-commit.ci
| self._fields_identifer(where), | ||
| self._fields_identifier(where), |
| if not isstr(custom): | ||
| custom = layout.parameter("__record__") | ||
| if not isstr(custom): | ||
| custom = layout.purelist_parameter("__record__") |
There was a problem hiding this comment.
Why was it once necessary to descend through lists to identify a record but now isn't?
I saw that there are some arrays of lists of records in the tests, which I'd think this is relevant for. If any of those tests override ufuncs, which I think they don't.
There was a problem hiding this comment.
Well, I tried it, and apparently this is fixing what I was thinking about, rather than breaking it.
In main:
>>> ak._v2.behavior[np.absolute, "Point"] = lambda data: np.sqrt(data.x**2 + data.y**2)
>>> array = ak._v2.Array([{"x": 1.0, "y": 1.1}, {"x": 2.0, "y": 2.2}], with_name="Point")
>>> np.absolute(array)
<Array [1.49, 2.97] type='2 * float64'>
>>> array2 = ak._v2.Array([[{"x": 1.0, "y": 1.1}], [], [{"x": 2.0, "y": 2.2}]], with_name="Point")
>>> np.absolute(array2)
Traceback (most recent call last):
...
File "/home/jpivarski/mambaforge/lib/python3.9/site-packages/awkward/_v2/_connect/numpy.py", line 109, in _array_ufunc_adjust_apply
out = apply_ufunc(ufunc, method, nextinputs, kwargs)
TypeError: <lambda>() takes 1 positional argument but 4 were givenbut in this branch:
>>> ak._v2.behavior[np.absolute, "Point"] = lambda data: np.sqrt(data.x**2 + data.y**2)
>>> array = ak._v2.Array([{"x": 1.0, "y": 1.1}, {"x": 2.0, "y": 2.2}], with_name="Point")
>>> np.absolute(array)
<Array [1.49, 2.97] type='2 * float64'>
>>> array2 = ak._v2.Array([[{"x": 1.0, "y": 1.1}], [], [{"x": 2.0, "y": 2.2}]], with_name="Point")
>>> np.absolute(array2)
<Array [[1.49], [], [2.97]] type='3 * var * float64'>So this must be the right thing to do. The thing I was worried about was it not finding the overloaded record when it's buried inside of a list like that.
But, uh-oh: there's another problem—if the array does not have the with_name="Point", ufuncs are not supposed to work on records. In v1:
>>> array = ak.Array([{"x": 1.0, "y": 1.1}, {"x": 2.0, "y": 2.2}])
>>> np.absolute(array)
Traceback (most recent call last):
...
File "/home/jpivarski/irishep/awkward-1.0/awkward/_util.py", line 1058, in apply
raise ValueError(
ValueError: cannot broadcast records in this type of operation
(https://github.com/scikit-hep/awkward-1.0/blob/1.9.0rc2/src/awkward/_util.py#L1060)(that's good), but in v2:
>>> array = ak._v2.Array([{"x": 1.0, "y": 1.1}, {"x": 2.0, "y": 2.2}])
>>> np.absolute(array)
<Array [{x: 1, y: 1.1}, {x: 2, ...}] type='2 * {x: float64, y: float64}'>(that's bad). But I tried this in main as well as this branch and it's the same in both, so it's unrelated to this change. I'll make it a new issue and prioritize it.
No description provided.