Re: Notes on the Metacity compositor

[Havoc Pennington <hp redhat com>]

BJörn Lindqvist wrote:
> That is just one example, but it seems to me that Metacity wasn't
> designed with flashy graphical effects in mind. One has to wonder: Is
> the effort required to make Metacity's compositing as good as Compiz
> greater or smaller than making Compiz as usable and mature as
> Metacity? Has anyone analyzed Compiz code and compared?

.02,

If the goal is no regressions to any metacity details (especially 
handling of bizarre in-house apps using 
motif/tk/plain-Xlib-and-GL/whatever, but also stuff like arcane ICCCM 
details, UI behavior details, a11y, and whatever), then starting from 
metacity is going to be easier. Joel's article applies,
http://www.joelonsoftware.com/articles/fog0000000069.html

*However* - "no regressions" is a dumb goal for most GNOME audiences. In 
fact, metacity became popular well before it had fixed all that gunge I 
just mentioned. Because most people never encounter or care about all 
that. (One reason it's hard to get it all fixed - the people who care 
about it are late adopters.)

In the end a "good enough" window manager just isn't that complicated. I 
was using Metacity for myself maybe a week or two into writing it. 
Looking at the ChangeLog, the first stable release was after 1 year, and 
while I may remember wrongly, I don't think I was working on it 
full-time. Of course, there are hundreds of fixes after that, but it was 
"good enough" after a year. A WM is not rocket science which is why 
there are so many of them.

So, if the goal is to get the new graphics capabilities, and be 90-95% 
OK on the details, then Compiz is unquestionably the faster route given 
that it's already there and people are working on it. And nobody is 
working on metacity compositing.

If Compiz _didn't_ exist then I wouldn't advocate starting from scratch. 
Some of the tricky bits from metacity could have been kept in a way that 
would avoid regressions, and I don't really agree with suggestions that 
the code could not be refactored. However, it would not have been as fun 
as starting over, or offered a sense of ownership/autonomy, so a 
volunteer developer probably wouldn't ever do this.

It kind of comes down to target audience again. The thin client and 
lots-of-old-UNIX-apps-in-the-enterprise kind of customers aren't going 
to see any value in Compiz, just pain from random regressions.

However, the Fedora/Ubuntu-using GNOME audience, which is probably 
larger, I'd expect to switch to Compiz pretty quickly, at least those of 
them who have an acceptable video card.

The FC6 solution (allow people to toggle metacity/compiz) seems like a 
good way to serve those two audiences, though the more compiz and 
metacity share the same interface to apps and the rest of the desktop 
(gconf keys, EWMH implementation, themes, default 
configuration/keybindings, etc.) the better it will work since 
documentation, apps, control panels, etc. won't have to be different 
across the two.

The hosed audience I'm guessing will be anyone who really needs just one 
or two simple compositing features (maybe a magnifier for a11y), but 
also needs to use some crufty old apps or a thin client setup or old 
video card or other weird cases that metacity already works with. The 
FC6 solution is no good for someone with that requirement set.

Maybe there's some value in making metacity a simple 2D compositor for 
this audience, I don't know.

The caveats for me personally about Compiz are

1) how much time people have spent on metacity, not by me in recent 
memory, but by quality contributors like Elijah; certainly not wasted 
time since metacity has been used by many people for many years, and 
will take a while to vanish yet, but it's always nice to see code live 
longer rather than shorter, and it would have been nice to keep some 
more of that work.

2) I don't think WM changes will bring GNOME to new audiences, it's a 
preaching to the converted kind of feature. With Vista and OS X, perhaps 
Compiz is like metacity was 5 years ago, in that it brings Linux up to a 
minimum standard of basic parity and sanity with the competition. But I 
don't think it creates an edge or killer app.

So if I had a wish it's that Compiz gets nailed down and "done" as 
quickly as possible, and people move on to things that will create an 
edge for _new_ audiences.


In the end Compiz is nice work and people like it and GNOME should take 
advantage.

And I can't tell you how much I'm looking forward to filing some 
completely unreasonable window manager bug and then raising hell when 
the maintainer rejects my demands ;-)

Havoc