I wonder on the use of the availability of so many similar but not quite identical modules

My rationale for writing Regexp::Assemble was that when I started looking around for something to do what I needed to be done, I didn't find what I wanted. Regex::PreSuf does not recognise metacharacters, so a\dz and a\sb produces something odd, like a\(?:dz|sb), which is uncompileable.

Regexp::Optimizer came closer to what I wanted, in that it understands metacharacters, but it doesn't do what I call tail folding. I have a large list of regular expressions that tend to share common tails and I was interested in producing the shortest expression possible (even though it can result in a more complex pattern that may actually perform a bit more slowly than if the tails were left alone).

For instance, on a sample of 4000 dictionary words weighing about 46Kb, the resulting R::A pattern is 23Kb long, whereas the R::O pattern is 28Kb long. For me the gain of 5Kb was an important consideration.

Another important issue is that while I consider Regexp::Assemble to be slow, it is no slouch. It assembles 3000 complex patterns into one in about 1.2 seconds. I fed the same set of patterns to Regexp::Optimizer, and 850 cpu seconds later it is still grinding away. (update: still chugging away, now at 11500 seconds. I am beginning to wonder whether it will terminate. 5Mb of core, so it's not swapping... later: 11 CPU hours later, Regexp::Optimize is still running. One may reasonably consider that it doesn't work on large patterns. There must be something exponential happening).

I look on in interest at demerphq's work in implementing tries in the regular expression motor, but as it doesn't deal with metacharacters at the moment, it's usefulness is somewhat limited for what I'm doing.