BUG: string replace results in invalid regular expression: invalid perl operator: (?<=

[jorisvandenbossche]

The pyjanitor clean_names method performs a bunch of regex .str.replace(..) calls, including with this regex, which is not supported by pyarrow:

>>> ser = pd.Series(['a@b', 'c@'], dtype='str')
>>> ser.str.replace('(?<=\\w)@(?=\\w)', '_', regex=True)
...
ArrowInvalid: Invalid regular expression: invalid perl operator: (?<=
>>> ser = ser.astype(pd.StringDtype(storage="python"))
>>> ser.str.replace('(?<=\\w)@(?=\\w)', '_', regex=True)
0    a_b
1     c@
dtype: string

[jorisvandenbossche]

I suppose this is actually a duplicate (or similar root cause) of #60833, which we forgot about, added back a milestone

[jorisvandenbossche]

Another invalid regular expression case, giving "invalid escape sequence: \Z"

>>> ser = pd.Series(['a', 'b'], dtype='str')
>>> ser.str.match('(?s:b.*r)\\Z')
...
ArrowInvalid: Invalid regular expression: invalid escape sequence: \Z
>>> ser = ser.astype(pd.StringDtype(storage="python"))
>>> ser.str.match('(?s:b.*r)\\Z')
0    False
1    False
dtype: boolean

Here, it seems that pyarrow does support \z but not \Z, so that is potentially something to replace in the regex.