Hi tye,

s/(?<=[0-9]{2})(?=[0-9]{2})/ /g
[download]

s/([0-9]{2})([0-9]{2})/$1 $2/g
[download]

I know that juxtaposing a look-behind and a look-ahead effectively matches a zero-width point between characters, for which you can then subsitute, as you did with a space. But is there a reason other than coolness to use that construct rather than the more familiar "match/capture/use captures in replacement" construct? Do you do so to avoid the overhead of populating and reading the capture variables?

Hello 1nickt,

I can’t answer for tye, but I think the following test script is instructive:

Read more... (804 Bytes)

Output:

 1:19 >perl 1307_SoPW.pl
                  Rate    capture lookaround
capture     87642419/s         --       -34%
lookaround 133333333/s        52%         --

 1:20 >
[download]

Hope that helps,

Athanasius <°(((>< contra mundum

Iustus alius egestas vitae, eros Piratica,

I would be tempted to go even a little further down the look-around path and specify that the four-digit groups be exactly four digits, i.e., that they be delimited by non-digits.

c:\@Work\Perl>perl -wMstrict -le
"my $s = '37 35 55555 37 8 17 16 25 36 37 3840 42 42 42 40';
 print qq{'$s'};
 ;;
 $s =~ s{ (?<= (?<! \d) [0-9]{2}) (?= [0-9]{2} (?! \d)) }{ }xmsg;
 print qq{'$s'};
"
'37 35 55555 37 8 17 16 25 36 37 3840 42 42 42 40'
'37 35 55555 37 8 17 16 25 36 37 38 40 42 42 42 40'
[download]

The OPed example data contain an instance of a single-digit data point: '37 8 17'. If I saw such a two-one-two pattern in the output, how would I know (other than by going back and examining the original data) that it was not a regex fix-up artifact?

c:\@Work\Perl>perl -wMstrict -le
"my $s = '37 35 55555 37 8 17 16 25 36 37 3840 42 42 42 40';
 print qq{'$s'};
 ;;
 $s =~ s{ (?<= [0-9]{2}) (?= [0-9]{2}) }{ }xmsg;
 print qq{'$s'};
"
'37 35 55555 37 8 17 16 25 36 37 3840 42 42 42 40'
'37 35 55 5 55 37 8 17 16 25 36 37 38 40 42 42 42 40'
[download]

Of course, there aren't any five-digit groups in the original data, right? Famous Last Words. This is, at least potentially, an illustration of the classic regex lament "I had a problem and I thought, 'I know, I'll solve it with a regex!' Now I have two problems." Much as I love regexes, I have to admit they can turn nasty on ya pretty quickly. Best to be precise.

Update: Come to think of it, if single-digit data values are valid, who's to say that the proper representation of '3840' is not '3 84 0'? More information is needed to make a useful decision here.

---

Give a man a fish:  <%-(-(-(-<