Here is complete story.

Thanks -- this was brief and clear enough that you should have put that into the OP, instead of making up a "pseudo/generic" case to confuse us. :)

Anyway, I would tend to prefer taking a two-step approach: apply all the substitutions that work, and after those are all done, if any residual template-variable markers are left behind, then croak. Something like this:

my %hash = ( var1 => 'value1', var2 => 'value2' );
  # ... or whatever...

while ( $tmpl =~ /@~(.*?)~@/ and exists( $hash{$1} )) {
    my $var = $1;
    $tmpl =~ s/@~$var~@/$hash{$var}/g;
}
croak "Bad template text:\n$tmpl" if ( $tmpl =~ /@~|~@/ );
[download]

"Basic" regex substitutions tend to be obscure and cryptic enough by themselves, without adding flow-control logic into the rhs. The code will be easier to maintain if you segregate the regex operations from the flow-control.