Seth, does this look similar to the regular expressions that trigger a
state explosion?
Hmm ... That is a good point. The pattern should be ok on its own, but
might be problematic in combination with others. It doesn't look like
there's much of a better way to detect this type, though, and it's
anchored to a specific byte at the beginning, which means it won't
kick in that often. Seth, any opinion on whether to take it out?
I suspect it’s probably ok. The biggest offenders are patterns that start with .*
Although I do remember that there was some issue with patterns that repeat a repeated body.
.Seth
I think it could become a problem if another similar pattern aligns
with this one on a DFA hot path. Ok, let's leave it in for now but
keep it in mind if we see trouble. At least the timing measurements
didn't show problems with the test-suite after I merged this (while I
think I recall indeed seeing improvements there when you did the DPD
regexp changes a while ago).
Robin
Yep, there were performance improvements after we fixed the regexp stuff before.
.Seth