Here is my solution to the common problem of parsing a format string typical of the formats used in log or printf specifications.

For example, to specify '[%t] [%l] %m' and have the parsed output be '[23:21] [err] Something bad happened.'

Simple string substitution is not enough, however. I need the substitution text to be the result of a function. :sigh: Actually, looking at my problem again, this isn't necessarily true, but I've been banging away at it under this assumption for so long, that I still want to see what other folks think.

my @tokens = split '', $ARGV[0];

my %meta = (
    't' => \&time,
    'l' => \&log_level,
    'm' => \&message,
);

my $max_iter = scalar @tokens;

TOKEN:
while ($indx < $max_iter) {

    if ($tokens[$indx] eq '%') {
        $indx++;

        for my $key (keys %meta) {
            if ($tokens[$indx] eq $key) {
                $output = $output . &{ $meta{$key} };
                next TOKEN;
            }
        }
        $output = $output . '%' . $tokens[$indx];
        next TOKEN;
    }
    $output = $output . $tokens[$indx];
}
continue {
    $indx++;
}

print "\n$output\n";
[download]

I just know that there must be a more elegant solution using substr, map, or s//, but all of these solutions eluded me.

---

In my regexp attempt, I stole from the Q&A node Subroutines in a regex? and tried this.

$format =~ s/(%m)/&message()/;
[download]

But this input 'the message is: %m' came out like this the message is: &message()

---

Conceptually, substr looked like a better fit, but I could not determine how to allow for the same metacharacter to be used more than once, even though there is not (yet) a valid reason for this.

---

The problem I ran into trying to use map was how to preserve the whitespace in the format string. If I didn't care about whitespace, I could have split the format string on whitespace and used map, but the whitespace is itself part of the format specification.