listcomp -> for-loop (listcopms create a new scope in py3), bypass Attri...

[metakermit]

...buteError on Pandas.str

Pushing pandas-dev/pandas#5530 upstream, so that the numpydoc version in Pandas can stay in sync.

[rkern]

👍

[pv]

Thanks, merged.

[stefanv]

As an educational exercise, would someone mind explaining this problem to me briefly?

[phaebz]

@stefanv This is a tricky issue with Python 3 class scope/namespace. I can point you to this and that SO question.

[stefanv]

@phaebz Thanks for that. It looks like both these examples deal with class namespace vs scope, though?

[pv]

I don't see issue in the list comprehension scoping as the iteration variable is not used afterward, but checking for the AttributeError is useful, and can't be done inside listcomp.

Py3 failure in the listcomp would have been caught by the test suite.

[stefanv]

Thanks, @pv, that makes sense to me.

[phaebz]

I can't reproduce the error, please just forget what I said.

[jorisvandenbossche]

There was a problem with pandas.Series.str returning an AttributeError, for which a fix (to return None) was applied at almost the same time as this fix was tried out. So probably the effect of both fixes were a bit mixed up.