Re: How to do this odd kind of regex match?

"Martin Ward" <Martin.Ward@durham.ac.uk> writes:
> more benefit from preprocessing the text. Build a hash table
> with the locations of all the 2, 3, 4 (or more) letter sequences.
> Then, to match against a regexp containing the word "porn"
> (say), you look up "porn" in the table and get the list of character
> offsets of locations of that 4 character string in the text.


This is essentially what the Perl RE engine does, by performing a FBM
analysis on the text, and then anchoring parts of a RE by doing an FBM
search for portions of the text. For instance, given /\w{3,5}foo/ and
"xxx abcdefoo", Perl does this:


floating `foo' at 3..5 (checking floating) stclass `ALNUM' minlen 6
Guessing start of match, REx `\w{3,5}foo' against `xxxx abcdefoo'...
Found floating substr `foo' at offset 10...
Starting position does not contradict /^/m...
Does not contradict STCLASS...
Guessed: match at offset 5


and starts hunting at the "b".


But I don't think that's what Tony's question was; from what I
understand it, that was "how do I store (potentially multiple levels
of) bracket-captured text when running multiple regexes in
parallel". I'm also working on a fast RE engine, and bracketed text is
where I'm coming unstuck as well, so if anyone's got a decent answer,
I'd love to hear it...
--
The debate rages on: Is Perl Bactrian or Dromedary?

Return to the comp.compilers page.
Search the comp.compilers archives again.