On Thu, Jun 18, 2009 at 07:26:58AM +0200, Michael Monnerie wrote:
> On Mittwoch 17 Juni 2009 Theo Van Dinter wrote:
> > Yes, it matters (one path is tried then the other has to be tried, as
> > opposed to having a single path)
>
> So which is better performance wise? I guess [sz]? but I'm not sure now.
It's very simple to test, here's a very unscientific one:
#!/usr/bin/perl
use Benchmark qw(cmpthese);
cmpthese(1000000, {
'group_nomatch' => sub { 'abc' =~ /b[xy]?/; },
'group_match' => sub { 'abc' =~ /b[xc]?/; },
'alt_nomatch' => sub { 'abc' =~ /b(?:x|y)?/; },
'alt_match' => sub { 'abc' =~ /b(?:x|c)?/; },
});
Rate alt_match group_nomatch group_match alt_nomatch
alt_match 534759/s -- -75% -75% -77%
group_nomatch 2127660/s 298% -- -2% -9%
group_match 2173913/s 307% 2% -- -7%
alt_nomatch 2325581/s 335% 9% 7% --
Not much difference, but alternation having a match ends up very slow.
Grouping would logically thinking be better and it seems so. Though it
always ends up slightly slower as a not matching alternation?
I'm sure there would be other test cases. Where are the perl über-geeks now? ;)
Cheers,
Henrik
