Re: Some perspective from the cheap seats...

Havoc Pennington writes:
 > 
 > To me the desktop people don't generally have an opinion about RENDER
 > on the level of detail that you're complaining about it. Most of us

Right. You shouldn't have to either.  It is our job to provide the
necessary abstraction so you are not exposed to these details.

 > will never use the render API directly. What we would want are high
 > level things like:
 >  
 >  - a way to alpha composite an image onto the current background
 >    without sucking the background off the server
 > 
 >  - a way to do vector graphics that produces the same results on both
 >    client and server side
 > 
 >  - a way to avoid redrawing the window manager frame and client window 
 >    in two different visible steps
 > 
 >  - a way to keep a magnified region of the screen up-to-date without 
 >    polling XGetImage() at some interval
 > 
 > In general we're fine with "fast enough" rather than "blazingly fast,"
 > and "a way" instead of "a perfect way," so sometimes software is

This may be a little short sighted. I would still encourage people
designing a new rendering model to give the hw people a consideration.

 > fine, sometimes somewhat odd APIs are fine. Oddness that we can bury
 > in the toolkit is generally not an issue, oddness that leaks out to
 > app developers is more of an issue, oddness that leaks out to users is 
 > a largish issue.

Yes. 
 > 
 > As I said earlier, most things that feel clunky right now are more
 > about client side performance or flicker/tearing/lack-of-animation
 > than about raw server performance.  Raw server performance is most
 > important to people doing 3D and other dedicated apps, rather than the
 > desktop. So performance isn't necessarily the primary issue for some
 > of this stuff.

Yes, the 'raw server performance' you are talking about was a stupid
marketing idea which forced everybody in a benchmark race that didn't
solve the realy issue namely the 'felt server performance'.
However flicker/tearing/lack-of-animation is very much an issue that
must be solved in HW.
The HW people need to make sure that all these features work together
seamlessly with things that are less in the focus of the UI community
like Xv and OpenGL. 

 > 
 > Usually the things we're asking for on the desktop are things Windows
 > or OS X already have, so they should at least be possible, I would
 > think. And for things that are only possible with *some* hardware,
 > feature tests can be fine too. (Just as you check for an extension.)

This is true. But to my knowledge (if anybody knows better please
correct me) todays drawing engines are designed so they can be interfaced
the Windows software APIs easily.
If ours are different interfacing is a lot harder.
 > 
 > In any case, the thing to do is probably to get more people involved
 > in new designs. What does the hardware look like? What did it look
 > like 3 years ago, what will it look like in 3 years? What API do we
 > want? How does all that interact? etc. Engineering is about tradeoffs,
 > just a matter of collecting all the data.

Right. Here we need people with knowledge in HW. A lot of things are
NDAed here so they cannot be discussed. 
Even if we don't like it we need to have a look at how DirectX does
things.

Egbert.