All excellent comments from JavaFan, BrowserUk, and eyepopslikeamosquito.

I hadn't considered the COMMIT/FAIL, partly because this came up when writing tests for some perl 5.8 code. It won't find overlapping matches, unlike the other tests. And I have to think about it more when used in the more general case. Still I'd like to use and try it out, so I'm adding it to the test suite.

As for BrowserUk's observation, yes character classes are preferable when looking for single characters, but this was meant for less trivial "atoms"... I will re-write the tests to be more realistic.

I'm glad that one of my solutions is exactly what The Perl Cookbook recommends, and the cookbook explains it better and talks about overlapping vs non-overlapping cases. Still, we have come up with some variations it does not cover.

New benchmark, taking the above into account, making the things to match/not match a little easier to modify:

#!/usr/bin/env perl
use Benchmark 'cmpthese';

my ($t1,$f,$t2)=(qr/people/,qr/babi?es/,qr/health(?:y|ful)/);

my %ways =
  (
   '3calls' => sub {/$t1/o && !/$f/o && /$t2/o},
   code     => sub {m:^(?(?{/$t1/o && !/$f/o && /$t2/o})|(?!)):},
   pos_look => sub {/(?=.*$t1)(?!.*$f)(?=.*$t2)/so},
   pos_anch => sub {/^(?=.*$t1)(?!.*$f)(?=.*$t2)/so},
   neg_look => sub {/(?!(?!.*$t1)|(?=.*$f)|(?!.*$t2))/so},
   neg_anch => sub {/^(?!(?!.*$t1)|(?=.*$f)|(?!.*$t2))/so},
   commit => sub {/$f(*COMMIT)(*FAIL)|^(?=.*$t1)(?=.*$t2).*$f(*COMMIT)
+(*FAIL)|(?=.*$t1)(?=.*$t2)/so}
  );

while (my ($way, $sub)=each %ways) {
  die "$way failed to match\n" unless ($_ = 'healthy people') && &$sub
+;
  die "$way had a false positive\n" if ($_= 'healthy people, including
+ babies') && &$sub;
}

print "For matching:\n";
$_="Our people eat heathful meals.";
cmpthese(-1, \%ways);


print "For non-matching:\n";
$_="Healthy people have more babes in arms";
cmpthese(-1, \%ways);
__END__
[download]

For matching:

	Rate	commit	code	neg_look	neg_anch	pos_look	pos_anch	3calls
commit	406208/s	--	-40%	-64%	-66%	-69%	-70%	-91%
code	675893/s	66%	--	-40%	-44%	-49%	-50%	-85%
neg_look	1128170/s	178%	67%	--	-7%	-14%	-16%	-75%
neg_anch	1212346/s	198%	79%	7%	--	-8%	-10%	-73%
pos_look	1312820/s	223%	94%	16%	8%	--	-2%	-71%
pos_anch	1344229/s	231%	99%	19%	11%	2%	--	-71%
3calls	4567476/s	1024%	576%	305%	277%	248%	240%	--

For non-matching:

	Rate	commit	code	neg_anch	neg_look	pos_look	pos_anch	3calls
commit	418623/s	--	-36%	-65%	-68%	-68%	-70%	-90%
code	654350/s	56%	--	-45%	-49%	-50%	-54%	-85%
neg_anch	1180978/s	182%	80%	--	-9%	-10%	-17%	-73%
neg_look	1294154/s	209%	98%	10%	--	-1%	-9%	-70%
pos_look	1308383/s	213%	100%	11%	1%	--	-8%	-70%
pos_anch	1418108/s	239%	117%	20%	10%	8%	--	-68%
3calls	4365151/s	943%	567%	270%	237%	234%	208%	--