you should test your regexp on ",foo," which will work, and shouldn't, as will "!!foo!!,".

So for a working single regexp you want (assuming \w is good enough)...

/^\w+           # Starts with a "word"
 (?:,\w+)*$/x;  # Followed by many "comma and word" atoms
[download]

Which IMO is ugglier than...

  /^[\w,]+$]/ &&  # Comma word atoms
! /^,|,,|,$/;     # With constraints on commas
[download]

It's probably faster to use 2 regexps too

you should test your regexp on ",foo," which will work, and shouldn't, as will "!!foo!!,".

It depends on what you think "should work". The OP's original regex was not anchored, and seemed intended to extract matching substrings rather than confirm that an entire string matched.

The /(\w+(?:\,\w+)*)/ regex will successfully extract the matching "foo" substring from your two sample cases into $1. If you want to check the entire string, then yes, leave out the parenthesis and use ^...$ anchors.

Update: with regard to It's probably faster to use 2 regexps too: Yes, a quick Benchmarking shows that, with anchoring, the double-regex style runs about 50% faster than the single-regex solution I posted. (Perhaps one of the resident RegEx gurus can explain why this is?)

However, if you want to extract matching substrings, I think the single regex is a sensible approach.

Update: with regard to It's probably faster to use 2 regexps too: Yes, a quick Benchmarking shows that, with anchoring, the double-regex style runs about 50% faster than the single-regex solution I posted. (Perhaps one of the resident RegEx gurus can explain why this is?)

Generally anything that looks like "(AB*)*" is bad for the backtracking.

--
James Antill

Update: with regard to It's probably faster to use 2 regexps too: Yes, a quick Benchmarking shows that, with anchoring, the double-regex style runs about 50% faster than the single-regex solution I posted. (Perhaps one of the resident RegEx gurus can explain why this is?)

I'd be interested to see your benchmark (code + data), as I don't come to the same conclusion. The benchmark below shows the one regex solution to be somewhat faster - the data sample is tiny though.

#!/usr/bin/perl

use strict;
use warnings;

use Benchmark qw /timethese cmpthese/;

chomp (our @lines = <DATA>);

our (@r1, @r2);

cmpthese -10 => {
    one      =>  '@r1 = map {/^\w+(?:,\w+)*$/          ? 1 : 0} @lines
+',
    two      =>  '@r2 = map {/^[\w,]+$/ && !/^,|,,|,$/ ? 1 : 0} @lines
+',
};

die "Unequal" unless "@r1" eq "@r2";


__DATA__
one,two,three,four,five
,one,two,three,four,five
one,two,three,four,five,
one,two,three,,four,five
one,two,three four,five




       Rate  two  one
two 25436/s   -- -26%
one 34417/s  35%   --
[download]

Abigail

\w will also match the underscore, which the orginal poster does not specifically require, and so should not be included.

Can someone explain this to me ?