While reading a thread on the Fun with Perl mailing list, an interesting question came about. Why does this code not produce a "use of uninitialized value" warning?

#!/usr/bin/perl -w

"abcde" =~ /(..)(?:(....)|(..))/;
# $1 is 'ab'
# $2 is undef
# $3 is 'cd'

print length($2);
[download]

Practically any other use of $2 would have resulted in the uninitialized value warning. And shouldn't it do so here? I mean, this code raises the warning:

#!/usr/bin/perl -w

print length($foo);
[download]

So why not with $2? The answer lies in the fact that the digit variables (and I think all punctuation variables) are magical, and as such their accesses are controlled by internal (C) functions. Specifically, length($2) calls the C function Perl_magic_len() (found in mg.c).

Examination of this function shows that variables which don't meet criteria to return a non-zero length end up falling through the cracks (intentionally) of the switch() statement, and end up returning 0. There's no code for raising a "use of uninitialized value" warning.

Why, you ask, does Perl_magic_len() not determine the length of the string returned from Perl_magic_get() (which governs fetching the variable's value)? Or, why doesn't length($2) call Perl's standard length() function with the value of $2? Because the variable is magic, and all of Perl's built-in functions (on the source-code level) have handles dealing with magic variables.

NB: one thing that irks me is that tied scalars do not have a LENGTH() method. Por que, Señor?

When length() encounters a magic variable as its argument, it calls that magic variable's length function. Normal scalars contain their length information as part of themselves (the SV structures in the C code). Magic variables are a completely different story. Chances are they don't actually hold any data at all. $2 gets its length from the source-code equivalent of the @- and @+ arrays, rx->startp[idx] and rx->endp[idx].

Paraphrasing the source code, when you ask for the length of $2, Perl does the following:

/* n is the number in $N */

if (
  (n <= rx->nparens) &&
  (s = rx->startp[n]) != -1 && /* -1 => not there */
  (e = rx->endp[n])   != -1
) return e - s;
return 0;
[download]

There's some added work for UTF8 in there, but I don't give a hoot.

There's your source-diving lesson for the day. I might do this from time to time, and not on topics as esoteric as this.

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