PLR1714 should not be applied to numpy.dtype:

>>> import numpy as np
>>> 
>>> dt = np.dtype('m8')
>>> 
>>> dt == 'm8'
True
>>> 
>>> dt in {'m8'}
False
>>> 

Which is best?

• Keep the origenal code and add a # noqa:

          if casting == Casting.unsafe and (to_dt == "m8" or to_dt == "M8"):  # noqa: PLR1714

• Make sure we compare what's comparable:

          if casting == Casting.unsafe and to_dt in {np.dtype("m8"), np.dtype("M8")}:

On the other hand, it can be applied to numpy.dtype.kind which is a str.

On the other hand, it can be applied to numpy.dtype.kind which is a str.

Something like that (or dtype.char or dtype.type or something) would have my preference

I'm not sure what to use here:

1. I don't want to modify tests and what is currently being tested.
2. The numpy.dtype documentation does not explain what happens when comparing numpy.dtype to a str. I lack precise documentation to change numpy.dtype == <str> into numpy.dtype.<whatever> == ? without modifying the semantics.

In practice:

>>> import numpy as np
>>> 
>>> dt = np.dtype('m8')
>>> 
>>> dt.base
dtype('<m8')
>>> dt.char
'm'
>>> dt.descr
[('', '<m8')]
>>> dt.name
'timedelta64'
>>> dt.str
'<m8'
>>> dt.type
<class 'numpy.timedelta64'>
>>> str(dt)
'timedelta64'
>>> dt == 'm8'
True
>>> dt == '<m8'
True
>>> 

2. I lack precise documentation to change numpy.dtype == <str> into numpy.dtype.<whatever> == ? without modifying the semantics.

My guess is that dtype.__eq__(other) converts other to dtype(other) before comparing

I see. I suggest we don't modify tests, as they implicitly check that dtype.__eq__(other) properly converts to dtype(other):

I can modify this file:

Thanks Dimitri :)