FURB118, `lambda x: x[:, 0]` should be `itemgetter((slice(None), 0))`, instead of `itemgetter((:, 0))`

[nasyxx]

In refurb, FURB118, lambda x: x[:, 0] should be itemgetter((slice(None), 0)), instead of itemgetter((:, 0))

# xx.py
import numpy as np
xs = np.array([[1, 2, 3]])
xm = map(lambda x: x[:, 0], xs)

xx.py:
  3:10 FURB118 [*] Use `operator.itemgetter((:, 0))` instead of defining a lambda

I guess (:, 0) is not a valid tuple.

[MichaReiser]

The python slice index syntax always trips me up and in this case it's unclear to me what the result of the map operation is.

@nasyxx could you clarify. Do you expect the above to be valid code? I'm asking to better understand if this is a bug in the fix or that the rule shouldn't flag the above code. CC: @AlexWaygood

[nasyxx]

@nasyxx could you clarify. Do you expect the above to be valid code?

Hi, the above is valid, while it will raise IndexError.

My actual ndarray is quite complex, you can change xs to xs = np.ones((5, 4, 3, 2, 1)). Then when running list(xm) will give you result.

[AlexWaygood]

There's definitely a serious bug here: we currently suggest invalid syntax (and, if you run with --fix, we introduce invalid syntax).

- lambda x: x[:, 0]
+ import operator
+ operator.itemgetter((:, 0))

In my opinion, itemgetter((slice(None), 0)) (which would be the correct way of rewriting this with itemgetter, as @nasyxx says) is quite ugly, so I think I'd vote for not emitting a diagnostic at all in this case. But others may feel differently; emitting a diagnostic would be consistent with the overall (quite opinionated) philosophy of FURB118.

[zanieb]

We already handle converting x[:] to slice(None) so I opted to continue emitting the diagnostic.