>>>>> "rm" == raymond meyer <raymond.meyer%rambler.ru@localhost> writes:
rm> I get the feeling that the Xorg developers don't care much
rm> about any other platforms that are not x86.
+1
rm> I also think X11 is too old, has too many hacks and too much
rm> bloat,
+1
rm> maybe it's time to move to something better in the future.
+1, but I think it's more likely we'll move on to something worse.
Everything new I can think of is myopically architected (ex., presumes
celfone screen, presumes no remote access at all), tiny and hacky and
``if only we could but we're stuck with _____'' (vnc), or is
disgustingly bug-riddled (freetype, xrender, gnome) and even more
bloated than X ever was so much that it tries to make bloat into a
virtue (``modularity'' and support for multiple ``platforms''). And
the best X server with the most ubiquitous and regression-free support
for a variety of extensions that are likely to become the core of a
future replacement architecture, is nVidia's closed source one.
the best case I can imagine would be moving the client-server pipe to
somewhere else in the software stack. for example there are some
nested X servers that offer the full protocol to clients, while
accessing the screen through a subset of the X protocol (some
``extension'') like DMX. This subset could become the new X protocol
over the network wire, and apps that depend on old Xlib functions
could start up one rootless X server per app to translate into the new
simpler protocol. First the rootless X server will be absorbed where
possible into these ``modularity'' blobs through the same mechanisms
said bloated stacks use to support ``platforms'' like Mac and win32,
and second the ``modularity'' blobs will start being swapped out:
someone will write a new blob that replaces a chain of three old
blobs. that sort of thing.
http://jonsmirl.googlepages.com/graphics.html
see ``three generations of windows''
he has basically the same idea---use rootless X servers to accomodate
old apps, and base the new application-to-screen interface on opengl.
he says using glitz/opengl behind Cairo instead of Xlib improves
performance 10 - 100x. so, use opengl as the new wire protocol,
along with some shims on either side of the wire to duplicate some
state and avoid round trips based on Chromium, but im not sure
exactly what Chromium is---I think it's useful as an idea not as code
but I'm not sure.
he also says XRender is basically a subset of opengl, which among
other things presumes the client will render the fonts, while it
might be smarter to still _store_ fonts on the client but have the
GPU render them so characters can be positioned sub-pixel. i dunno
about this. maybe.
so far on Unix it seems like what fonts are installed and selected is
the overwhelming issue, and the new ways of rendering them are doing
more harm than good because shitty fonts often get chosen in web
browsers---no one is keeping track of it end-to-end so now and then
the font choices are good, while most unixes most times are in the
midst of a prolonged regression with a bunch of out-of-date howto's
for getting yourself partway out of the regression. a bigger standard
collection of higher quality free fonts, and a way for app developers
to avoid using fonts outside the collection by accident, even if they
were the most primitive sort of .pcf, would be worth far, far more to
real users than all this fun bickering about how to render the fonts.
I'm close to the point of ditching an entire Unix (Gentoo in favor of
Ubuntu) just because it is easier than groveling through all the tiny
font packages and baroque xml font-choosing config files and mountains
of freetype2 bug reports to get good fonts in the damn web browser,
only to find some other program still rendering 5-pixel-high uglyfonts
in dialog boxes. It highlights the real disaster of X---all this
argument about the future elegant architecture is irrelevant when
getting keyboard layouts and mouse buttons and fonts to work in the
most basic way is the overwhelmingly hardest part of setting up the
whole system. You get it almost working and then find out some
jackoff foreign keyboard program is grabbing control-space, and some
retarded start-button window manager is grabbing ALT-<a buncvhh of
windows shortcuts> and making emacs useless. and what the fuck
happened to cutting and pasting from xterms which hasn't worked right
in over a decade: things are always getting silently unselected, or
something will be obviously still highlighted right in front of you
while middle-button pastes some older cut buffer. In 1997 this used
to work every time perfectly. bloated or not as the old stuff was
this new crew is a bunch of monkeys.
Attachment:
pgpLel2vRGPJ_.pgp
Description: PGP signature