[cells-gtk-devel] Re: Cells-gtk now UFFI-ized, running under AllegroCL and Lispworks on win32, next stop OS/X.

[Andras Simon]

[resending now that I'm subscribed to the list]
[to Kenny: commenting out the listbox works now, so I left out the part 
complaining about it]

On Wed, 22 Dec 2004, Kenny Tilton wrote:

>
>
> Andras Simon wrote:
>
> >
> > > > I'll send you a tarball later today. It'll have everything but the
> > > > tree demo working. But, because of that, an announcement may be just
> > > > a bit too early.
> > >
> > >
> > > Ah, yes. That tree demo was a monster. I actually did that in two
> > > steps, because it itself has two sub-demos: "listbox" and "treebox".
> > > So I commented out one until I got the other working. Each took days.
> > > :( Hope what I did carries over to CMU.
> >
> >
> > It definitely did! Both parts of the tree demo work out of the box; it's
> > just that they're so bloody slow that I gave up before the window came
> > up. But now I put 1 in the :init-value slot of the h-scale, and that
> > helped.
>
> Nice guess! How slow was "slow"? I have seen like a 3-4 second pause, no
> more. Never investigated.
>
> If your "slow" is much out of line with the Clisp original (using CLisp
> FFI -- did you ever try building that to see if it would fly?) or if it s
> out of line with ACL and Lispworks on win32, I will suspect the CMUCL FFI
> interacting with UFFI, a known problem.

It's very much out of line with ACL. I don't have numbers, but it must be
something like thousands of times slower.

> As for Cells being the problem, one never knows, but the original pure
> CLisp version uses Cells, and Cells are pretty solid as a rule.

I thought that maybe cells/CMUCL was to blame. The funny thing is that all
the other demos are pretty fast. Perhaps it's time to take a hard look at
the compilation notes and warnings. (A CMUCL expert would come in handy.)

>
> I am going to spend some time merging any changes I made ot get a clean
> compile under LW on Mac OS/X and then update CVS with everything. Then off
> to cll to emtion the CMUCL coverage.

Given that the tree view demo makes the startup time for the whole demo
unacceptably long, I suggest #-cmu-ing out "Tree-view" in test-gtk.lisp
before going public.

Andras