ENH: Change implementation of pandas.api.types.is_list_like to check for iterability in more pythonic way

[dcoukos]

Is your feature request related to a problem?

Yes, this would improve predictability of behavior with pandas extension arrays.

Describe the solution you'd like

pandas.api.types.is_list_like can check for iterability in a more pythonic way according to https://docs.python.org/3/glossary.html#term-iterable. This would result in the appropriate call being iter(obj) rather than checking for __iter__ attribute.

API breaking implications

This will not affect the API.

Describe alternatives you've considered

Alternative solutions include communicating to third parties that they must change their extension arrays to include hasattr(q, "__iter__") == False to make sure that classes with ambiguous data iterability (may contain either scalar or array-like data) are perceived as utterable when they contain scalar data.

Additional context

Opened a bug report for this issue. I think an enhancement request is more appropriate, but I am linking the previous (closed) bug report as it includes useful information.
#43195

[dcoukos]

Looks like this has actually been discussed earlier, even though I had not found them when searching.

Not sure if development of a solution for this problem has stalled.
For reference:

#35127
#35131

hgrecco/pint#1128

[jbrockmendel]

cc @burnpanck closed by #43196?

[burnpanck]

While not implementing the proposed enhancement exactly, it does enable the feature motivating this request to be implemented. Therefore, it's fine with me if this is closed.

The enhancement remains unimplemented, as is_list_like still only does an approximate model of an iterability check. It only checks if the type supports iterability in general, but not if the specific value does. Before #43196, it special-cased numpy arrays by taking the dimensionality of the value into account. Now, this special-case is extended to some duck-arrays (specifically, any value having a .ndim attribute).

However, is_list_like never intended to actually model iterability exactly - it explicitly special-cases str and bytes as not list-like despite being iterable. Therefore that would require a fundamental API discussion - what should Pandas even consider as list-like? Is it really iterability with the exception of str and bytes? Then, implementing this would cause further performance concerns.

[jbrockmendel]

@dcoukos closing this as solved, pls reopen if im wrong on that point.