develooper Front page | perl.perl5.porters | Postings from March 2000


Thread Previous | Thread Next
Ilya Zakharevich
March 28, 2000 16:01
Message ID:
M.J.T. Guy writes:
> > In the symbol table, the same way as overload subroutines, with name
> > similar to ")body".
> Yukk!    I'd regard kludgy names like as much worse than another
> special subroutine name.     But that's a matter of taste.

Why?  You do not know they exist until you explicitly scan the symbol
table.  And you are not supposed to scan it until you are ready for

Propose (and implement! ;-) a better scheme for inheritable overloading.

> > No.  In my style the interpreter knows which thing is provided, and
> > which is not.
> Why does the interpreter need to know?    If you don't want one
> of the functions, just make your subroutine return 'undef' in that
> case.

There are several (related) reasons, none of them decisive when
considered separately.

  a) With your scheme a reference to a subroutine is returned, even if
     only a prototype is needed.  Thus you either insert a subroutine
     in a stash, or create/destruct a fake subroutine.  Both solutions
     are resource hogs, one for memory, another for speed.

  b) Even if you do not want to do compile-time loading, with your
     scheme a subroutine call is performed for each unknown symbol (hog).

  c) Your scheme does not allow coexistence of compile-time loading
     with AUTOLOAD (may be desirable for incremental improvement of
     the existing module).


Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About