So these are really "on_next_writable" and "on_next_readable" rather than general event handlers. (Unless they are reset with again().) That's definitely different from POE, in which event handlers persist once set.

So, conceptually, it's sort of like this (in a sort of pseudo-code and with some lexical scope differences so that the function references are explicit which I think makes the order of execution clearer):

my ($req, $socket, $buf);

sub talk {
  $req = shift;
  $socket = IO::Socket::INET-> new( PeerAddr => 'www.perl.com', PeerPo
+rt => 80);
  return lambda \&init;
}
  
sub init {
  when_writeable( $socket, \&write_handler );
}

sub write_handler {
  send_request( $req => $socket );
  when_readable( $socket, \&read_handler );
}

sub read_handler {
  read_stuff( $socket => $buf ) or die;
  if ( done() ) {
    return $buf;
  } else {
    when_readable( $socket, \&read_handler );
  }
}  

my $q = talk( HTTP::Request-> new( GET => 'http://www.perl.com') );
print $q->wait;
[download]

That's definitely different from POE, in which event handlers persist once set.

Quite correct. I didn't realize that this is a key point difference until you've noticed it - thank you! I agree, it surely is. I should emphasize that in the docs.

And yes, the code does the right flow.