I know of 3 main reasons to want to reverse a regexp. Firstly it swaps "can't have variable-length lookbehinds" for "can't have variable length lookaheads", and sometimes a variable-length lookbehind is really useful. Secondly, tail-anchored variable-length patterns are difficult to optimise (for much the same reasons as make variable-length lookbehind difficult to support), and reversing makes them head-anchored and optimisable. Thirdly it is an interesting intellectual pursuit that may tell us more about the mathematics of the patterns we express through regexps (and that might in turn supply new insights which would benefit us in unanticipable ways).
Probably the simplest common example is stripping trailing whitespace with s/\s+\z//, an incantation often liberally sprinkled around people's code. Because there is no specific optimisation that handles this, the runtime will increase with the number of embedded spaces in the text:
#!/usr/bin/perl -w
use Benchmark;
our $nonspace = ('abc' x 100) . "\n";
our $space = ('a c' x 100) . "\n";
Benchmark::cmpthese(-1, {
space => q{ my $copy = $space; $copy =~ s/\s+\z// },
nonspace => q{ my $copy = $nonspace; $copy =~ s/\s+\z// },
});
Rate space nonspace
space 32880/s -- -89%
nonspace 300755/s 815% --
Hugo |