Re-reading history from 2001 / using a capture during split

[talexb]

I've been part of this community for a while, and it's always fun to check out Selected Best Nodes for pearls (heh) of wisdom. Today I found Being pretentious, and getting away with it., which talked about the behaviour of split with a particular regex.

I started writing a module to beautify code recently, and ended up using a capture in a split statement -- which isn't something I do regularly. Here's the interesting part about this subject (finally) .. the example in the node talked about the following behaviour:

In Perl 5.005_02, split /(A)|B/, "1A2B3" returned a five-element list of (1, 'A', 2, undef, 3). In 5.005_03, it returned (1, 'A', 2, '', 3); a subtle, but meaningful, difference. There's only one way to get an undef from split(), and that's from the underlying regex match. A capturing paren that does not match has undef as its $DIGIT value.

A tiny bit of incorrectly written code in pp.c:pp_split() caused these undefs to become empty strings. Bah. I corrected it, documented it, and tested it.

Cool, I thought, and ran off to test it in the debugger.

  DB<2> @foo = split /(A)|B/, "1A2B3"

  DB<3> x @foo
0  1
1  'A'
2  2
3  undef
4  3
  DB<4>

Oh.

The behaviour has gone back to the previous one. I believe it's because split is splitting on the un-captured 'B', and since it's not captured, undef is the correct result, rather than a null string (with apologies to japhy.

Thoughts?

EDIT: Updated title to add '/ using a capture during split' to the title, as per feedback.

[vr]

split is splitting on the un-captured 'B', and since it's not captured...

No, it's the (A) that's not captured. Every group produces a field, as last paragraph in split explains.

If the PATTERN contains capturing groups, then for each separator, an additional field is produced for each substring captured by a group (in the order in which the groups are specified, as per backreferences); if any group does not match, then it captures the undef value instead of a substring

  DB<3> @foo = split /(A)|(B)/, "1A2B3"


  DB<4> x @foo
0  1
1  'A'
2  undef
3  2
4  undef
5  'B'
6  3

I'm reading linked node as japhy apologizes for what he did (i.e. making "split" to return empty string), and acknowledges that returning "undef" was correct from the very begining.

[choroba]

Yes, and to get rid of the undefs, one can use the branch reset pattern:

my @foo = split /(?|(A)|(B))/, "1A2B3";
use Data::Dumper; print Dumper \@foo;

$VAR1 = [
          '1',
          'A',
          '2',
          'B',
          '3'
        ];