[oe] [RFC] Xserver-common 1.30

[Stanislav Brabec]

Marcin Juszkiewicz wrote:
> Hi
> 
> We have two packages which do similar things: 
> 
> - xserver-common from GPE
> - xserver-kdrive-common from Poky
> 
> Basically they do the same -- select proper XServer, arguments to pass 
> for it and have session scripts. Kdrive version uses 'xinit' to 
> initialize X11 session as root user, GPE version is expected to be used 
> with gpe-dm/gpe-login combo which handles user login.
> 
> This gave us few mistakes in past - some developers patched one of them 
> for their device but the other one was used in image etc.

Yes semi-pseudo-universal files makes it hardly maintainable.
Maintainers of different platforms afraid or removing obsolete hacks for
a different platform and add their own hack instead.

> Some time ago I looked at xserver-common 1.25 and created over 20 
> patches for it:
> 
> - moved xmodmap files to xmodmap/ dir
> - use machine_id from init.d/functions

What about /etc/default/machine_id (possibly generated on the first
boot)? It would save several launching of AWK interpreter in scripts.

> - xserver:

Well, universal simple wrapper using machine-specific
/etc/default/xserver would make sense. I can imagine arguments like:
XSERVER_TYPE, KDRIVE_EXTRA_ARGS, XORG_EXTRA_ARGS, DPI,
DISPLAY_ORIENTATION, INPUT_ORIENTATION and others, if they will be
needed.

In particular, Xserver script tries to cover both xorg and kdrive, but
xorg/kdrive switch is installed-packages-based, but machine detection is
hardware based.

As a result, installing xorg on a machine configured for kdrive causes a
Xorg syntax error - xorg does not understand most of the arguments, just
only -dpi.

> - xserver: introduced MOUSE variable for "-mouse" argument (KDrive only)
Or KDRIVE_MOUSE= in /etc/default/display?

> Most of patches were used to create 1.30 version which is available as 
> tarball from GPE project. But due to my fault this version is broken...
> 
> Today I pushed 1.30 into OE metadata with few other patches which are 
> expected to fix errors left in 1.30 release. It was not too much tested 
> on devices so beware if your target do not works properly with it.
> 
> In short my patching ideas were:
> 
> - set DPI in separate variable instead of being part of ARGS
>   + some devices needs only DPI setting
>   + default ARGS are usually fine for devices
> - use /etc/default/xserver for setting ARGS/DPI for new devices
>   + no need to patch Xserver anymore
>   + easy to tweak on device itself

Yes, it's a good idea.

> - xmodmap stuff should be machine dependent so only one xmodmap is
>   present on device

Yes, and skip checks for platform in 12keymap

>   (this part was not done yet)

Agree, but:

- the keymap-fixup multiple load hack should be moved to kdrive >= 1.4
packages.

- XKB-less kdrive installations may even live completely without X
keyboard map

- Xorg installations may live without calling xmodmap explicitly

Other notes:

setdpi looks like a black magic, making X resources file machine
specific would make more sense

40xmodmap. What is the purpose? Which platform has /proc/hal?

01xrandr (set the default default orientation) seems to be obsolete

> - zaurusd, devmand should provide session scripts instead of being
>   called from Ts_calibrate script
>   (this part was not done yet)

zaurusd should be started during boot. There is no need to start it in X
session scripts (well, the way zaurusd accesses DISPLAY is incredibly
ugly). I don't know devmand.

> I would like other developers to use it and test does it work on their 
> devices.
> 
> In future I would like to add support for using "xinit" in 
> /etc/X11/Xserver so 'xserver-common' package could be used for rootmode 
> X11 systems (which would eliminate 'xserver-kdrive-common' package).
> 
> What do you think about it?

Yes, everything that gets rid mix of common code, platform specific
hacks, hardware specific hacks and software specific hacks makes a lot
of sense.

-- 
Stanislav Brabec
http://www.penguin.cz/~utx/zaurus