Like I already told you in the CB:

my $re = join "|", map { quotemeta $_ } keys %entitylist;
$re = qr/(?:$re)/;

# ...

$string =~ s/($re)/$entitylist{ $1 }/ge;
[download]

You might want to take a look at perlre for the /e switch.

Is there a reason for the non-capturing parens in "$re = qr/(?:$re)/;"?

What good does the /e switch do in the regex "s/($re)/$entitylist{ $1 }/ge;"?

Anno

No (qr// already adds non-capturing parens) and none (interpolation is done after each match).

The /e switch is an error by me - it is a leftover from when I thought about doing the hex-conversion manually with a sprintf call in the right hand side..

I always build my regular expressions with noncapturing parentheses when I pal on latter assembling them - this prevents embarassing bug hunts later when I change how the target RE is built, possibly by repeating one ("atomic") building block - leaving out the parentheses causes hard-to-track misbehaviour with input on the seam of the two blocks.

.

Off the top of my head:

my $entity = join '|', keys %entitylist;
$string =~ s/($entity)/$entitylist{$1}/g;
[download]

This way you only scan the whole string once instead of once for each entity.

Update: I've gotten a private message that this needs a /e switch to work. I submit that this is not the case.

my %entitylist = ( a => 1, b => 2);
my $string = 'abc';
my $entity = join '|', keys %entitylist;
$string =~ s/($entity)/$entitylist{$1}/g;
print $string, "\n";
__END__
12c
[download]

(I actually tested this before my original post, but I only pasted in the relevant portion.)

Do the contents of the file have to be in one $string?

If you can read the lines of the file into an array instead, it should take less processing time, as the regex won't have to scan (and modify) one huge, single line each time.

It's unescaped HTML entities:

I want to convert the entities like &ge, &le, into hexa values &#x2265, &#x2264.

But, I'll also point out that the actual entity should end in a semicolon. (which prevents issues such as '∨' matching 'ª')

Well then, it's something like below that might fit.


Read more... (2 kB)
The results are interesting:

Benchmark: timing 1000 iterations of regexish, with_splitting...
  regexish: 59 wallclock secs (58.16 usr +  0.02 sys = 58.18 CPU) @ 17
+.19/s (n=1000)
with_splitting:  0 wallclock secs ( 0.02 usr +  0.00 sys =  0.02 CPU) 
+@ 50000.00/s (n=1000)
            (warning: too few iterations for a reliable count)
[download]