Backreferences in regexes match, but do not capture

Question

This program

say "zipi zape" ~~ /(\w)\w» \s+ \w+({$0})/;

returns

｢pi zape｣
 0 => ｢p｣
 1 => ｢｣

which I interpret as the backreference to the first match being matched to a zero-width match? Maybe because it's matched to $0, which is itemized to '' outside the regex? How could I use these backreferences, and capture at the same time the match? Note: this related to this documentation issue, which requires clarification of the use of backreferences.

At the time of writing this all the issues raised and questions asked in your original question **and** in your comments on Håkon's answer are carefully explained in [my answer to "Why/how is an additional variable needed in matching repeated arbitary character with capture groups?"](https://stackoverflow.com/a/56397290/1077672). Do you think it's worth walking thru your version of the problem and questions? I will happily write an answer if you read my linked answer and think yours is different but for now I think it's essentially a duplicate. — raiph, Jul 07 '19 at 12:10

score 4 · Answer 1 · answered Jul 07 '19 at 10:09

4

According to the documentation:

If you need to refer to a capture from within another capture, store it in a variable first

So you could use:

say "zipi zape" ~~ /(\w){} :my $c = $0; \w » \s+ \w+($c)/;

Output:

｢pi zap｣
 0 => ｢p｣
 1 => ｢p｣

answered Jul 07 '19 at 10:09

Håkon Hægland

39,012
21
81
174

Actually, I wanted to understand above behavior. But there's an additional thing I kind of not-understand; why do you use open and closing braces behind the first capture? – jjmerelo Jul 07 '19 at 10:15
1

It seems this usage of `{}` is only mentioned in a [comment](https://github.com/perl6/doc/blob/master/doc/Language/regexes.pod6#L1088): it says we need it to update the current match object. – Håkon Hægland Jul 07 '19 at 10:23
1

Hi @jjmerelo "I wanted to understand above behavior." In brief, `(...)` is shorthand for creating a new `Match` object and setting `$/` to that new object inside the parens. So your second pair of parens means that the `$0` inside them does *not* refer back to the (capture from the) first pair of parens. – raiph Jul 07 '19 at 12:21
1

@jjmerelo "why do you use open and closing braces behind the first capture" I've explained this (both in summary and in great depth) in prior SOs; see my comment on your question as a starting point. – raiph Jul 07 '19 at 12:22
@raiph Thanks for the reference to the other question! – Håkon Hægland Jul 07 '19 at 12:31

score 3 · Accepted Answer · answered Jul 07 '19 at 13:40

{$0} isn't a backreference.
It is a code block.
In this case, it is a code block which does absolutely nothing.

In order for it to actually be used as a part of the regex it needs <> around it.

In fact since () denotes something like a new closure in respect of $/, it would be an empty regex if it was actually being used for something.
($/ is reset for every (), so $0 is also reset.)

say "zipi zape" ~~ /(\w)\w» \s+ \w+(<{$0}>)/;

Cannot resolve caller INTERPOLATE_ASSERTION(Match:D: Nil:U, BOOTInt, BOOTInt, BOOTInt, BOOTInt, PseudoStash:D); none of these signatures match:
    (Match: Associative:D, $, $, $, $, $, *%_)
    (Match: Iterable:D \var, int \im, int \monkey, int \s, $, \context, *%_)
    (Match: Mu:D \var, int \im, int \monkey, $, $, \context, *%_)
  in block <unit> at <unknown file> line 1

That happens because it is basically the same as (<{Nil}>).

What you could do is update $/ before the second () by using {}, and use double quotes around $0

say "zipi zape" ~~ /(\w){}\w» \s+ \w+("$0")/;
｢pi zap｣
 0 => ｢p｣
 1 => ｢p｣

Too me this seems a little unreliable.
(It is relying on what I would consider a mis-feature, if not an outright bug.)

This is where we get to Håkon Hægland's answer of storing it in a lexical variable.
(After updating $/ by using {}.)

say "zipi zape" ~~ /(\w){} :my $c = $0; \w » \s+ \w+($c)/;

Lexical variables are not scoped to (), so it is perfectly safe to do this.

I would personally stringify $0 since that is the only part of the match object inside of $0 that is being used.

say "zipi zape" ~~ /(\w){} :my $c = ~$0; \w » \s+ \w+($c)/;

Honestly I don't see a reason to even capture the second match, since it is always going to be the same as the first match.

say "zipi zape" ~~ /(\w)\w» \s+ \w+$0/;

I also see little point in adding » since the \s+ already forces it to be the end of a word.

say "zipi zape" ~~ /(\w)\w \s+ \w+$0/;

Backreferences in regexes match, but do not capture

2 Answers2