Aside from building one combined regex by using Regexp::Assemble, it comes down to a tradeoff between doing multiple substr() queries against the string and running one regex against the string. If you care, the development version of Perl, "bleadperl", has an improvements regex engine by demerphq which make such searching faster.

The only optimization besides using a different version of Perl that could make your regex faster would be to collect all the first letters of all your words and make them into a zero-width lookahead character class I think, but I would make sure by benchmarking that this really brings a speedup.

I was inspired to benchmark your advice. In my testing environment, I was astonished to find out that the speed increases as you collect more and more letters (first letters, then second letters, etc.) in your look-ahead.

Benchmark: timing 200 iterations of lookahead1, lookahead3, lookahead4
+, lookahead5, nolookahead...
lookahead1:  3 wallclock secs ( 2.93 usr +  0.00 sys =  2.93 CPU) @ 68
+.26/s (n=200)
lookahead3:  2 wallclock secs ( 2.23 usr +  0.00 sys =  2.23 CPU) @ 89
+.69/s (n=200)
lookahead4:  2 wallclock secs ( 2.03 usr +  0.00 sys =  2.03 CPU) @ 98
+.52/s (n=200)
lookahead5:  2 wallclock secs ( 2.00 usr +  0.00 sys =  2.00 CPU) @ 10
+0.00/s (n=200)
nolookahead:  7 wallclock secs ( 6.30 usr +  0.00 sys =  6.30 CPU) @ 3
+1.75/s (n=200)
[download]

Read more... (2 kB)