ENH: set __module__ on IndexSlice

[shengjie2013]

• Part of ENH: set __module__ on top-level public objects #55178
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

[simonjayhawkins]

@jorisvandenbossche this one is a bit different?

[simonjayhawkins]

Thanks @shengjie2013 for the PR

[simonjayhawkins]

not sure about this....

>>> pandas.IndexSlice
<pandas._IndexSlice object at 0x7ff1e9305bd0>
>>> pandas._IndexSlice
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'pandas' has no attribute '_IndexSlice'. Did you mean: 'IndexSlice'?
>>> ```

since the purpose of #55178 is that the repr should be the path end users access the objects. This is an odd case.

[jorisvandenbossche]

The actual IndexSlice object is defined a bit lower as IndexSlice = _IndexSlice(), so we would need to update the __module__ of that object, I think

[jorisvandenbossche]

See

pandas/pandas/core/indexing.py

Line 148 in d3c595e

IndexSlice = _IndexSlice()

[jorisvandenbossche]

Although, for a class instance, that doesn't actually seem to work:

In [3]: pd.IndexSlice.__module__ = "pandas"

In [4]: pd.IndexSlice
Out[4]: <pandas.core.indexing._IndexSlice at 0x7fa7d01c3610>

In this case we might need to add a def __repr__ to the _IndexSlice class if we want to improve this.

[simonjayhawkins]

so we would need to update the __module__ of that object, I think

so it looks like the way this would be done is instance.__class__.__module__ = 'new_module_name' which changes the module attribute for the class which is what has been done anyway.

the output for a class is of the form <class 'pandas.core.indexing._IndexSlice'>

whereas the output for a class instance is of the form <pandas.core.indexing._IndexSlice object at 0x7fda1576c280>

now changing the __module__ attribute on the class (as done already in this PR) gives <pandas._IndexSlice object at 0x7f6442deff70> which is not correct since `_IndexSlice is not a top level class.

I'm assuming that if we want to change this, we would want <pandas.IndexSlice object at 0x7f6442deff70>? This would be technically incorrect also?

In this case we might need to add a def repr to the _IndexSlice class if we want to improve this.

in another discussion it was mentioned that the repr is not being used in IPython?

It appears that changing the __name__ attribute on the _IndexSlice class (or alias) doesn't get the desired result either.

[jorisvandenbossche]

in another discussion it was mentioned that the repr is not being used in IPython?

I think that was about function objects, but here we have an instance of a class, and that should use its repr (that's how we define repr of other pandas objects)

But it seems you have found another solution anyway

[simonjayhawkins]

i'll try something that does give the correct output and is technically correct but we'll see what else breaks!

>>> pd.IndexSlice
<pandas.IndexSlice object at 0x7f2528c70d60>

[simonjayhawkins]

i'll try something that does give the correct output and is technically correct

in jupyter notebook the output is <pandas.IndexSlice at 0x7f2d681cd870>

but we'll see what else breaks!

seems fine so far except for the pre-commit error

[jorisvandenbossche]

Wondering if we should keep a _IndexSlice = IndexSlice here for backwards compatibility (I know this is private, but you never know ..)

[jorisvandenbossche]

From a quick github search I don't find much public use, but for example the stubs use the current import to get the class for type annotations

[simonjayhawkins]

no problem. can do that if you are happy with the rest of it.

[simonjayhawkins]

From a quick github search I don't find much public use, but for example the stubs use the current import to get the class for type annotations

hmm. I'll look in more detail (probs tomorrow now) before making the changes since if the class is needed for type annotations then perhaps the class should be exposed in pandas.api.typing and the repr should point to that.