You know, I think you might have nailed it. Intuitively that seems correct.

I did it this way to divide the parsing of the input file from the format of the output. So the backend would say "I've found a glark element", and frontend A may say "Ignore it", while frontend B may say "Concatenate it with the preceding output".

The plan was for the client to say:

my $obj = Backend->new(frontend=>FrontendClass);
[download]

But, as you point out, far more sensible is for them to say:

my $obj = Frontend->new;
[download]

Then to have Frontend use Backend. :-) Thank you very much!

Yep, that works perfectly. I had a feeling it was a fundamental mistake. :-)

I now use Frontend directly like this:

  my $obj = Frontend->new;
  $obj->parse($file);
[download]

Frontend inherits from Backend, and Frontend doesn't provide a _parse_ method, so it dispatches it up the chain to Backend. Backend does have this method and starts parsing the file. When it hits a token it calls

$self->handle(name=>'token_name',args=>['arg1','arg2'])
[download]

which is a generic dispatch mechanism in Backend.

The code for _handle_ is simplified now because it does something like:

if ($self->can($token)) {
  return $self->$token($args);
}
[download]

IOW, it looks for a method with the same name as the token in the current class and when it fails it passes it to Frontend which does implement said method.:-)