Re: List-to-Range generation

[ZZamboni]

OK, it took me a few minutes to completely understand how and why this works. Here is my dissected version of the regex:

     s/(\d+)            # first number (group #1)
        (?:             # group #2
         ,              #   followed by a comma
         (              #   group #3
          (??{$++1})    #     match previous number + 1 (group 4)
         )              #   end group #3
        )+              # end group #4, repeat
       /$1-$+/gx;       # substitute for the first number followed by 
+the
                        # last matched one

Group #1 matches the first number in a sequence of numbers. Then, the ??{$+ + 1} is used to match "the last number plus one" ($+ stands for whatever was matched by the last set of grouping parenthesis). For the second number in a sequence, the "last number" is the one matched by group #1. But for subsequent numbers (because of the +), the last number matched (this is, whatever the ??{$++1} matched last time) becomes the "last number". So the thing repeats until the "last number plus one" part doesn't match anymore (this is, until a non-consecutive number is found), and then replaces the whole thing with the first number (group #1), a dash, and the last number matched.

At first look, I thought the double parenthesis around ??{$++1} were unnecessary, but without them it does not work, and here is why: $+ contains what was matched by the last set of parenthesis, not the current set. So by doubling the parenthesis, it makes $+ contain the last thing matched by the current expression. Very clever!

--ZZamboni

[japhy]

Almost.

m{
  (\d+)           # \1 start -- digits -- \1 end
  (?:
    ,             # ,
    (             # \2 start
      (??{$++1})  # evaluate '$+ + 1' as a regex
    )+            # \2 end (and try again)
  )
}

The $+ refers to the last successful captured pattern, and that capture must have been closed. So the first time the (??{...}) is reached, $+ is $1's value. The next time, it's $2's (first) value, and then $2's new value, and so on.

japhy -- Perl and Regex Hacker

[rmocster]

Any chance to structure it so to list padded numbers. I like to convert (0001,0002,0003,011,012,013,015) to "0001-0003,011-013,015". Thanks a lot.

[hippo]

Our venerable learned brother ZZamboni graces our humble monastery with his esteemed presence only infrequently these days. He last visited some 18 months ago, so you might be waiting a while for a direct reply.

Other interested parties might wish to know that rmocster subsequently posted his own SoPW question (Convert an array of numbers into a range). You might wish therefore to follow that thread to see not only the context but the ensuing discussion.