yegg has asked for the wisdom of the Perl Monks concerning the following question: 


#!/usr/bin/perl

my @strings = ('testing','','testing','testing');

foreach my $string (@strings) {

    print "\nSTRING: $string\n";

    my $string2 = 'http';
    $string2 =~ /^h(.*)/o;

    my $q = $string;
    my $test = 'testing';
    warn $test =~ /\Q$q\E/;
}

[download]

I finally got to the bottom of the problem I reported at http://www.perlmonks.org/?node_id=867059 (thx to everyone who commented), and I distilled it to the above test case.

Here's what I think is happening. If you have a an /o regex with a capture clause and it captures something, and then you run a a second regex with a variable, but it set to '', then it uses the first compiled regex instead (and thereafter), and doesn't match as you'd expect as a result.

In this example, You'd expect line 14 to warn 1 then nothing, then 1 twice. Instead, you get 1, then nothing the rest of the time.

I'm running v5.8.9 on FreeBSD 7.0. FWIW, I also filed a perlbug here: http://rt.perl.org/rt3/Ticket/Display.html?id=78564

Replies are listed 'Best First'.
Re: Perl bug or feature?
by Corion (Patriarch) on Oct 25, 2010 at 17:10 UTC

    It's not a bug in Perl but a bug in your expectations. The empty RE // always matches, see perlre resp. perlop. And the first match has nothing to do with it, as removing it still produces all matches:

    #!/usr/bin/perl

my @strings = ('testing','','testing','testing');

foreach my $string (@strings) {
    my $q = $string;
    my $test = 'testing';
    warn "$test =~ /$q/?";
    warn $test =~ /\Q$q\E/;
}

__END__
testing =~ /testing/? at tmp.pl line 8.
1 at tmp.pl line 9.
testing =~ //? at tmp.pl line 8.
1 at tmp.pl line 9.
testing =~ /testing/? at tmp.pl line 8.
1 at tmp.pl line 9.
testing =~ /testing/? at tmp.pl line 8.
1 at tmp.pl line 9.

    [download]

    Update: Actually, while my reply is not untrue, the real cause is told by Anonymous Monk below. The empty match will repeat the last match.

[reply]
[d/l]
[select]
      Actually, the empty regexp, // is special. See perlre, resp. perlop. 
      say "foo" =~ /oo/ ? "MATCH" : "NO MATCH";
say "bar" =~ // ? "MATCH" : "NO MATCH";
MATCH
NO MATCH

      [download]
      An empty pattern repeats the last successful match.
[reply]
[d/l]
[select]
Re: Perl bug or feature?
by Anonymous Monk on Oct 25, 2010 at 17:19 UTC
[reply]
      Even though I knew about — and respect — perl's behaviour to repeat the match with an empty regex //, I do feel that this should not apply to the case where the empty regex is a result of interpolating an empty variable: $re = ''; /$re/

      Usually, the variable's data comes from user input, and that should not fall back to a shortcut which is intended for intra-source only.

[reply]
[d/l]
[select]
Re: Perl bug or feature?
by JavaFan (Canon) on Oct 25, 2010 at 17:27 UTC
    I think this is indeed a bug. In fact, it's very easy to see it's a bug - the /o on the first match is redundant (as there's no variable in the pattern), yet removing the /o changes the outcome.

    I can confirm this bug is present in 5.12.2 as well.

    Running the code with re debugging clearly shows the bug: 

    Compiling REx "^h(.*)"
Final program:
   1: BOL (2)
   2: EXACT <h> (4)
   4: OPEN1 (6)
   6:   STAR (8)
   7:     REG_ANY (0)
   8: CLOSE1 (10)
  10: END (0)
anchored "h" at 0 (checking anchored) anchored(BOL) minlen 1
Guessing start of match in sv for REx "^h(.*)" against "http"
Guessed: match at offset 0
Matching REx "^h(.*)" against "http"
   0 <> <http>               |  1:BOL(2)
   0 <> <http>               |  2:EXACT <h>(4)
   1 <h> <ttp>               |  4:OPEN1(6)
   1 <h> <ttp>               |  6:STAR(8)
                                  REG_ANY can match 3 times out of 214
+7483647...
   4 <http> <>               |  8:  CLOSE1(10)
   4 <http> <>               | 10:  END(0)
Match successful!
'testing' =~ /testing/ at w line 14.
Compiling REx "testing"
Final program:
   1: EXACT <testing> (4)
   4: END (0)
anchored "testing" at 0 (checking anchored isall) minlen 7
Guessing start of match in sv for REx "testing" against "testing"
Found anchored substr "testing" at offset 0...
Guessed: match at offset 0
1 at w line 15.
Guessing start of match in sv for REx "^h(.*)" against "http"
Guessed: match at offset 0
Matching REx "^h(.*)" against "http"
   0 <> <http>               |  1:BOL(2)
   0 <> <http>               |  2:EXACT <h>(4)
   1 <h> <ttp>               |  4:OPEN1(6)
   1 <h> <ttp>               |  6:STAR(8)
                                  REG_ANY can match 3 times out of 214
+7483647...
   4 <http> <>               |  8:  CLOSE1(10)
   4 <http> <>               | 10:  END(0)
Match successful!
'testing' =~ // at w line 14.
Freeing REx: "testing"
Compiling REx ""
Final program:
   1: NOTHING (2)
   2: END (0)
minlen 0
Guessing start of match in sv for REx "^h(.*)" against "testing"
String not equal...
Match rejected by optimizer
Warning: something's wrong at w line 15.
Guessing start of match in sv for REx "^h(.*)" against "http"
Guessed: match at offset 0
Matching REx "^h(.*)" against "http"
   0 <> <http>               |  1:BOL(2)
   0 <> <http>               |  2:EXACT <h>(4)
   1 <h> <ttp>               |  4:OPEN1(6)
   1 <h> <ttp>               |  6:STAR(8)
                                  REG_ANY can match 3 times out of 214
+7483647...
   4 <http> <>               |  8:  CLOSE1(10)
   4 <http> <>               | 10:  END(0)
Match successful!
'testing' =~ /testing/ at w line 14.
Guessing start of match in sv for REx "^h(.*)" against "testing"
String not equal...
Match rejected by optimizer
Warning: something's wrong at w line 15.  
Guessing start of match in sv for REx "^h(.*)" against "http"
Guessed: match at offset 0
Matching REx "^h(.*)" against "http"
   0 <> <http>               |  1:BOL(2)  
   0 <> <http>               |  2:EXACT <h>(4)
   1 <h> <ttp>               |  4:OPEN1(6)
   1 <h> <ttp>               |  6:STAR(8)
                                  REG_ANY can match 3 times out of 214
+7483647...
   4 <http> <>               |  8:  CLOSE1(10)
   4 <http> <>               | 10:  END(0)
Match successful!
'testing' =~ /testing/ at w line 14.
Guessing start of match in sv for REx "^h(.*)" against "testing"
String not equal...
Match rejected by optimizer
Warning: something's wrong at w line 15.
Freeing REx: "^h(.*)"
Freeing REx: ""

    [download]
    There's no way the last match should be /^h(.*)/.
[reply]
[d/l]
[select]
Re: Perl bug or feature?
by AnomalousMonk (Archbishop) on Oct 25, 2010 at 19:33 UTC

    With either
        $string2 =~ /^h(.*)/o;
    or
        $string2 =~ /^h(.*)/;
    (i.e., either with or without the  /o regex modifier), I get the same results for ActiveState 5.8.9 and Strawberries 5.10.1 and 5.12.0 (all running under Windoze 7) and as I would expect (after a little thought) given the documented behavior of the  // empty regex: 

    >perl -wMstrict -le
"my @strings = ('testing','','testing','testing');
 foreach my $string (@strings) {
     print \"\nSTRING: $string\n\";
     my $string2 = 'http';
     $string2 =~ /^h(.*)/o;
     my $q = $string;
     my $test = 'testing';
     warn $test =~ /\Q$q\E/;
 }
"

STRING: testing

1 at -e line 1.

STRING:

Warning: something's wrong at -e line 1.

STRING: testing

1 at -e line 1.

STRING: testing

1 at -e line 1.

    [download]
[reply]
[d/l]
[select]
Re: Perl bug or feature?
by ig (Vicar) on Oct 26, 2010 at 04:51 UTC

    While initially I found it surprising, after reviewing Quote and Quote like Operators and a few tests with use re qw(debug All), I came to the conclusion that it is merely not obvious and wouldn't be inclined to call it a bug. Some examples in perlop to clarify what empty means might be helpful, and sufficient from my point of view.

[reply]

Back to Seekers of Perl Wisdom