With all of the benchmarks flying around, I don't see anyone testing the original suggestion of /this|that|etc/ without the trie fanciness.

      nada:   0.39 CPU @  2.56/s (n=1)
      trie:   3.73 CPU @  0.27/s (n=1)
        or:   6.04 CPU @  0.17/s (n=1)
    alt_re:   8.52 CPU @  0.12/s (n=1)
   alt_sub:   8.62 CPU @  0.12/s (n=1)
     index:   9.17 CPU @  0.11/s (n=1)
      list:  15.38 CPU @  0.07/s (n=1)
[download]

Note that "or" is faster than either "alt" version, which was my original point. I was just curious is that was still true since I couldn't find the documentation that stated that any more.

"nada" just reads the data. "trie" is tilly's stuff. "or" is my sub with lots of little regexen. "alt_re" is one big regex with lots of alternates compiled with the qr// operator. "alt_sub" is the same but wrapped in a subroutine and compiled with eval. "index" is my looping and using index(). "list" is looping using lots of little regexec compiled with qr//.

I think "index" and "list" lose here because of all of the looping for non-matching lines.

        - tye (but my friends call me "Tye")