[FEA] Regex: Provide option for '.' to not match '\r'

[andygrove]

Is your feature request related to a problem? Please describe.
In Python and cuDF, the default behavior is that . will match all characters except for line-feeds, so this will match on \r. However, the default behavior in Java (as used by Spark) is that . will not match \r so this leads to an incompatibility.

cuDF example:

>>> cudf.Series(['a\rb']).str.contains('a.', regex=True)
0    True
dtype: bool

>>> cudf.Series(['a\nb']).str.contains('a.', regex=True)
0    False
dtype: bool

Spark example:

scala> val df = Seq("a\rb", "a\nb").toDF("c0")
scala> df.createOrReplaceTempView("t1")
scala> spark.sql("select c0 rlike 'a.' from t1").show
+-----------+                                                                   
|c0 RLIKE a.|
+-----------+
|      false|
|      false|
+-----------+

Describe the solution you'd like
I would like an option to specify whether . should match \r so that we can match the default Java behavior?

Describe alternatives you've considered
None

Additional context
None

[davidwendt]

I'm reluctant to add specific support for \r character since it could create confusion in other places where \n is handled -- like for ^ and $ as well as builtin classes \W and \D.
Could you replace the \r with \n in the input string? Since both are single byte characters, this could be done in-place on the chars column.

[andygrove]

I think that the best I can do in terms of a workaround is to inspect the input first to see if it does contain \r if the regex pattern contains . and fall back to the CPU in that case. Replacing \r with \n could have side effects but it all depends what the regex pattern is.

[davidwendt]

Another option is to use [^\r\n] instead of .
This would match any character but \r or \n.

[andygrove]

@davidwendt Thanks for the suggestion. We have gone with the approach of transpiling the regular expressions and replacing . with [^\r\n] so I am closing this issue,