Fwd: Re: tigert's mockups and HTML

[Mike Hearn]

> Let's face it, there are no more desktops that don't have an html renderer
> *already* loaded.  We should focus our efforts on making it so that each
> *instance* of the renderer doesn't suck tons of memory, not on avoiding HTML
> based on the rather suspicious theory that we can avoid loading an HTML
> renderer.

The per-instance overhead of a rendering engine is fairly high, at 
minimum you have (in gecko at least):

- the input text/images/css

- the render content tree. I'm trying to remember if the DOM is merged
   with the internal trees in Gecko, I seem to remember that at one point
   they were separate and then merged (or vice-versa)

   This can be quite big, as it's a set of complex objects with a
   lot of associated state, such as computed style.

- the output pixmap

That's a lot more than a few widgets!

Also just because Mozilla/KHTML is loaded in the background doesn't mean 
it's paged in.

> In fact, if we use HTML for more onscreen stuff, we can probably save
> memory: a lot of the rendering we would have to code by hand (ie. layout of
> messages in a notification window) could be done in HTML.

No, I am pretty confident that Gecko/KHTML would use more memory than 
GTK or Qt to render a couple of images/labels simply because it's doing 
more work. Also it's more likely that a widget toolkit is paged in than 
a rendering engine.

thanks -mike