False positive: Unmatchable dollar in regular expression with lookahead assertion

[sam-maloney]

Description of the false positive

I get false positives for the following expression (using the Python re module):

r"^(?=\[.*]$|[^][]*$)\[?[1-9]\d*(\+|-[1-9]\d*)(:[1-9]\d*(:[+*^])?)?]?$"

Unmatchable dollar in regular expression
This regular expression includes an unmatchable dollar at offset 9.
This regular expression includes an unmatchable dollar at offset 17.

It seems to be complaining about the dollars in the forward lookahead assertion, i.e. (?=\[.*]$|[^][]*$), but both of those occur at the end of their respective alternatives, so they can match fine, and because it's just a lookahead assertion it doesn't actually consume the match and the following expression can still be tested as normal.

Code samples or links to source code

The regex has been developed/tested here: https://regex101.com/r/G4MP54

[smowton]

Minified example: re.compile(r"^(?=a$)a$") triggers a false-positive, presumably because the lookahead group is followed by something non-nullable.

[smowton]

@tausbn has posted a prospective solution at #20495

[smowton]

Fix will appear in CodeQL 2.23.2, ETA 2 weeks.