The solution becomes clear when one realizes that qr// regexes are nothing but strings (albiet special string), and that they are not interpolated until runtime. Consider:

    my $inlinetag = do {
        my $acc =
        qr/(?> \{          (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+  \} )/x;
        qr/    \{ (\w+): ( (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+) \}  /x;
    };
[download]

In this modified snippet, the $acc regex is constructed, and scoped lexically to the do block. Next, the unnamed regex is constructed. Note that both of these pass strict checking since $acc was created in the same block, and thus exists. Since the second regex was the last statement in the do block, it is returned, and is assigned to $inlinetag. I believe this is the behavior you expected.

So what is the problem? The problem is that when $inlinetag is expanded at runtime, $acc no longer exists since it went out of scope when the do block returned the unnamed regex and assigned it to $inlinetag. A potential solution is just to get rid of the do block:

    my $acc =
        qr/(?> \{          (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+  \} )/x;
    my $inlinetag =
        qr/    \{ (\w+): ( (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+) \}  /x;
[download]

Or, if you're really worried about cluttering your namespace, you can do some closure magic:

    my $inlinetag = sub {
        my $acc =
        qr/(?> \{          (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+  \} )/x;
        qr/    \{ (\w+): ( (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+) \}  /x;
    };
[download]

However, you must now use it like this: qr/(??{$inlinetag->()}/ rather than like qr/$inlinetag/ with the first method.

Wonderful explanation. Thanks!

- Yes, I reinvent wheels.
- Spam: Visit eurotraQ.

Would a closure be close to what you want?

{
    my $acc = '';
    sub inlinetag {
#   ... whatever this does
    }
}
[download]

A lexical is not constructed until the my statement declaring it hits its semicolon.

After Compline,
Zaxo

Would a closure be close to what you want?

Not really, as the snippet I pasted is part of a larger whole:

{
    my %exlinx;
    my $link       = qr/ \[ (?: ([^|\]]*) \| )? ([^\]]*) \] /x; # 1: u
+rl 2: text
    my $inlinetag = do {
        our $acc =
        qr/(?> \{          (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+  \} )/x;
        qr/    \{ (\w+): ( (?: (?> \\. | [^\\{}]+ | (??{ $acc }) ) )* 
+) \}  /x;
    }; # 1: tag 2: text
    my $overig     = qr/(?: (?> \\. | [^{\[\\]+ ) )*/sx; # -: text
    my $stuff      = qr/(?> $overig ) | (?> $link ) | (?> $inlinetag )
+/x;
    my $paratag    = qr/(\w+):\s*/;
    my $n          = qr/\n\s*\n/;

    # These return html-ified versions of the input
    sub inlineparse { ... }
    sub paraparse { ... }
    sub colorize { ... }
}
[download]

it may help to note that (??{}) are done like quoted eval, compiled in an interlude during runtime.

That does help a little, but why doesn't it give me the famous Global symbol "$acc" requires explicit package name, then? I have no $acc declared in the scope where $inlinetag is being used.

- Yes, I reinvent wheels.
- Spam: Visit eurotraQ.