RE2
This is not really an issue but a general thing to consider. Since the parsers use regular expressions a lot, it would make sense to optimize this part. It seems that the re2 library which ímplements a non-backtracking regular expression engine could be significantly faster than the default re-library that used currently. Also switching could be as easy as:
import re2 as re
But I'm sure there are possible problems with missing functionality. From what I understand the python library will also fall back to the default implementation whenever back-tracking is needed in the matching. When we have the time to optimize things a bit, we could benchmark it.