This is a cool idea! I'm not sure if my tactic is much better but I thought of using a source filter:

package CFGPL;

use warnings;
use strict;

our $Comment_Start = '#';

sub import
{
    shift;
    $Comment_Start = quotemeta shift if @_;

    # Read in only our commented code...
    open my $own_src_file, '<', $0;
    my @perl_code = grep { /^$Comment_Start/ } <$own_src_file>;
    close $own_src_file;

    die "Config file has no code in comments\n" unless @perl_code;

    # Write it back to config file and erase uncommented stuff!
    open STDOUT, '>', $0;
    print @perl_code;
}

# Filter out the leading comment and runs the code...
require Filter::Simple;
Filter::Simple::FILTER( sub { s/^$Comment_Start//gm; } );

1;
[download]

.XCompose-ordering has the same format as .XCompose. (I include the former from the latter.) It consists of lines of usually the following format:

<XKeySym>(?: <XKeySym>)* : (?:"string"|U[\da-fA-F]{4})(?: description)?

String is usually a single UTF-8 char, but is a C-style string (\n \r \\, etc.). And Uxxxx represents a single Unicode char. The format also supports '#'-comments, which last to the end of the line, ala perl. And it supports 'include' lines:

include "other file"

Full specs

Can you include a literal newline inside the double quotes? That would allow you to use perl -x:

<not> <gonna> <happen> : "
#!perl

.
. This is the script
.

__END__
"

.
. This is the actual file
.
[download]

So what would the script look like? How about

local $^I = '~';
local @ARGV = __FILE__;

my $seen_data;

while (<>) {
    if (!$seen_data) {
        $seen_data = $_ eq qq{__END__\n};
        next;
    }

    ... change $_ as desired ...
}
continue
{
    print;
}
[download]