Re: Standalone Pango, take two

[Torsten Schoenfeld <kaffeetisch gmx de>]

muppet wrote:
> > • Performance: 'use Gtk2' got 3% slower on my machine.  (That is, 'use 
> > Gtk2' now takes 1.03 times the time it takes without the patch.)  
> > Should we worry about that?
>
> That's not very much, and it's undoubtedly due to the added seeks & DSO 
> load for Pango.so, as well as the API fixups.
>
> The obvious fix would be lazy-loading, which might actually be a win for 
> lots of programs.  After all, you don't *have* to call Pango functions 
> in your user code to have a useful Gtk2 app.
>
> The trick would be figuring out how to defer loading Pango.pm (and the 
> corresponding Pango.so) until actually needed.  That starts to sound 
> like more complexity than warranted for fixing a 3% slowdown.

Yeah, I agree.

> >  perl -MPango -le'map { print } keys %Pango::' | sort
> >
> > More importantly even, the loop will also automatically map over any 
> > new API we add to Pango, thus indirectly changing Gtk2's API.  Is it 
> > better to hardcode the stuff that should be mapped over?
>
> That's not a feature?  ;-)
>
> The hardcoded list would effectively freeze the Gtk2::Pango API, which 
> might be a good idea.

I converted the loop to use a hardcoded list now.

> > • The yellow-wiring foreach loop somehow makes these two evaluate to 
> > true:
> >
> >  Pango::Layout->isa (Gtk2::Pango::Layout::)
> >  Gtk2::Pango::Layout->isa (Pango::Layout::)
> >
> > I don't understand why or how it does that.  Is that a mechanism we 
> > can rely on?  It is important that the two expressions are true for 
> > backwards compatibility.  (I tested it with perl 5.8.8, 5.10.0, and 
> > blead -- it seems to work fine in every case.)
>
>
> If i'm reading the globbing expression correctly, it looks like putting 
> the same references in both symbol tables instead of just copying.

Yes, and somehow this makes the isa() algorithm think that both packages are 
equivalent.  I asked about this on the module-authors list: 
<http://www.mail-archive.com/module-authors perl org/msg06814.html>.  The mail 
also contains a small example program that uses no modules.

> s/relevent/relevant/g in the pod.

Fixed.

-Torsten