BUG: .extractall() throws AssertionError if capture group length > 1

[GeraintDuck]

Code to replicate error:

import pandas as pd
s = pd.Series(["a13a23", "b13", "c13"], index=["A", "B", "C"])
s.str.extractall("[ab](\d\d)")

Note that the regex [ab](\d) from the documentation page works, whereas [ab](\d\d) above doesn't. It seems that any captured group with a length of > 1 causes this error.

Though playing with this a bit more, the following regex's all seem to work correctly without error:

([ab])(\d\d)
()[ab](\d+)
(a13)(\d\d)

I've reproduced the issue in both versions 0.18.0 and 0.18.1. I'll admit I've not checked against the master branch though.

Note: I posted this to the mailing list, but haven't had any responses - thus I assume this is a bug.
I'm unsure what the underlying cause is here (maybe it doesn't like the first regex character not being within a capture group?).

[jreback]

cc @tdhock

yeah this looks buggy. welcome to have you step thru and see where the error is / pull-request.

[GeraintDuck]

I've narrowed it down to the list comprehension on this line. When the tuple is of length 1, it separates out the string into individual chars, rather than iterating over the (single) tuple value(s).

>>> [x for x in ('ab', 'xy')]
['ab', 'xy']
>>> [x for x in ('ab')]
['a', 'b']

I suppose code can be added beforehand to check for this particular case (though checking the length has the same issue - both examples above are length 2), but is there a cleaner fix?

Checking the total number of groups in the regex match might be an alternative.

[jreback]

you can just make a list of the capture groups i think

In [9]: [ x for x in ('ab')]
Out[9]: ['a', 'b']

In [10]: [ x for x in ['ab']]
Out[10]: ['ab']