But i dont think this will scale very well... (and probably has subtle problems anyway)

One quibble is that because \d is a subset of \w then a string such as "abc123def" will get \w{9} in your version. Here's a slightly improved version (for some definition of improved)

my $string=" \aabc123def!*#\n";
$string=~s{  ([[:digit:]]+)
            |([[:alpha:]]+)
            |([[:punct:]]+)
            |([[:space:]]+)
            |([[:cntrl:]]+)
            |(.)
          }
          {
              defined($1) ? '[[:digit:]]{'.length($1).'}'
                       : defined($2) ? '[[:alpha:]]{'.length($2).'}'
                       : defined($3) ? '[[:punct:]]{'.length($3).'}'
                       : defined($4) ? '[[:space:]]{'.length($4).'}'
                       : defined($5) ? '[[:cntrl:]]{'.length($5).'}'
                       : "\Q$+\E"  # anything else?
                                     
          }gex;
print $string;
[download]

But it still has problems (for example, \n is in both :space: and :cntrl: so "\n\a" produces [[:space:]]{1}[[:cntrl:]]{1}, but "\a\n" produces [[:cntrl:]]{2}).

One quibble is that because \d is a subset of \w then a string such as "abc123def" will get \w{9} in your version.

Yup. But personally I consider that a feature not a bug. :-) After all ldkjdlkjf2098kklls probably isnt [[:alpha:]]+\d+[[:alpha:]]+

But we are both in agreement that there isnt a good way to do this, although as we both have shown there are a variety of bad ways to do it... BTW, is the . really necessary? I dont think it is as the s/// will just skip the char if it doesnt match.

Oh and I considered using something like you post here, but I fgured that considering I tend not to use the POSIX char classes that much probably others wouldnt either.

:-)

Yves / DeMerphq
---
Software Engineering is Programming when you can't. -- E. W. Dijkstra (RIP)

But we are both in agreement that there isnt a good way to do this

Agreed. Anything that tries to generalize beyond "\Q$string\E" requires a variety of assumptions.

BTW, is the . really necessary? I dont think it is as the s/// will just skip the char if it doesnt match.

Just trying to be careful :-) In case I neglected something with those classes, and that something also required escaping, then just leaving it in the string wouldn't result in a valid re. (I don't tend to use POSIX char classes either, so I wasn't sure exactly how inclusive I was being).

After all ldkjdlkjf2098kklls probably isnt [[:alpha:]]+\d+[[:al­pha:]]+

I did play around once with a small script that did these things, when I was bored, and no, it probably isn't. But what I did was take several strings and tried to derive a common expression out of them - first looking for similarities, like a sequence of numbers in the middle, or whitespace at the end, or whatever, and then built sub-regexes from the parts. I think starting with splitting on non-words and such gave so-so results for things like email addresses..

Of course, I never really got any really usable results, but it was a fun exercise. :) What I wanted to say was that do decide which it should be you need a decent sample of several strings that should all match. Then it is sometimes possible to get something to build upon. Maybe. :)

---

You have moved into a dark place.
It is pitch black. You are likely to be eaten by a grue.