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

We've jut upgraded from 5.4 to 5.6.1, and tripped over what appears to be a bug: specifically, ".*?\322" as the search target of a regular expression doesn't work (actually, it's any highbit character). 
perl -w -e '$_="Hello there\n"; s/t/\322/; print $_; s/^.*?\322/yes/; 
+print $_'
Hello ̉here
Hello ̉here

[download]
Clearly,in the second line, the correct output is "yeshere". I have determined that "(.*?)\322", "[^\322]*" and ".*\322" all work (but the last obviously means something slightly different). Since we've got this problem in existing code, I'd rather find a patch, or a mechanism for finding the problem areas, rather than having to go through all of our code, checking by eye. Any ideas where to start?

Edit Masem 2001-09-14 - Code tags in examples

Replies are listed 'Best First'.
Re: High bit bug in RE?
by japhy (Canon) on Sep 14, 2001 at 17:45 UTC
    Got it! I had to tell C to compare the characters as unsigned char instead of char.

    _____________________________________________________
    Jeff[japhy]Pinyan: Perl, regex, and perl hacker.
    s++=END;++y(;-P)}y js++=;shajsj<++y(p-q)}?print:??;

[reply]
      Glad to hear it. But, um... does that help me, or just the next version of Perl?
[reply]
        You could recompile Perl if you want if you make the adjustment to the source code. Or you could change the way the regex is written.

        _____________________________________________________
        Jeff[japhy]Pinyan: Perl, regex, and perl hacker.
        s++=END;++y(;-P)}y js++=;shajsj<++y(p-q)}?print:??;

[reply]
Re: High bit bug in RE?
by Zaxo (Archbishop) on Sep 14, 2001 at 13:01 UTC

    That should be \0322. The leading zero makes it an octal number. \322 matches a previously captured match.

    Update: Hmmm... I dropped the ball on regex numbers. This seems to do what you want: 

    $ perl -w -e '$_="Hello there\n"; s/t/\322/; print; s/^.*\322/yes/; pr
+int'
Hello ̉here
yeshere

    [download]
    The minimalizing '?' seems to be the problem. Enclosing (.*?) in parens also works.

    After Compline,
    Zaxo

[reply]
[d/l]
      I'm afraid that that doesn't make any difference; neither does using \xD2 instead. There's a section in the perlre documentation which implies \322 only means a previously captured match if I've got 322 brackets (although I'm possibly misreading it).
      Further, using  $a=chr(210); s/.*?$a/yes/; doesn't work either.
      However, all the problems go away if I use \177 (ie. 127 decimal) or lower. So there's a definate inconsistency. :-(
[reply]
[d/l]
        I'm trying to fix this right now. As for why putting the .*? in parens worked, that's because of a failure to optimize that I've now fixed.

        _____________________________________________________
        Jeff[japhy]Pinyan: Perl, regex, and perl hacker.
        s++=END;++y(;-P)}y js++=;shajsj<++y(p-q)}?print:??;

[reply]

Back to Seekers of Perl Wisdom