Re: Profiling guile-gtk

[Andy Wingo]

Yo,

On Thu, 2004-05-13 at 22:50, Andy Wingo wrote:
> Where does this lead us? The obvious target is method definition. There
> are two approaches to this: make the method registration go faster, and
> delay the registration until later.

I implemented delayed generic binding. This reduces (gnome gtk) load
time on my machine to about 4.4 seconds. Just loading (gnome gw gtk) is
3.8 seconds or so. If we implement delayed types, this figure might go
down to 3.5 seconds or so for the whole library. Better, but still not
good. I turned GC back on, BTW. Turning it off made it churn a lot when
it was finally allowed. The proper solution to that is not to allocate
so much in the first place.

Attached is the patch to g-wrap--tng/guile/g-wrap/guile-runtime.c. Note
the the cache for generics had to be put in the root module so as to
allow different modules to cache methods on the same delayed generic.
Thus it's pretty hackish.

We need to get the load time down to about a second or so, I'd say.

By the way, I can't get guile-gtk-demo to work with the new g-wrap. It
uses the pack-start generic, which I think is having problems with
gtk-box-pack-start's optional arguments.

Regards,
-- 
Andy Wingo <address@hidden>

guile-runtime.c.diff
Description: Text Data