Comment 8 for bug 1217194

The trick, perhaps, is to *actually* use "true" regexes - that is, regular (in the technical sense of the word) expressions. An NFA implementation typically has reasonable performance. The problem in the example arises when the parentheses (used for syntactic grouping only in REs) are overloaded with support for 'match groups' in the output. Do we ever need to get at these subgroups within heat?

If this is to support an "allowed pattern" only, then finding an RE library that turns off backref/match group support and falls back to a stock NFA implementation would be one potential approach.