[MPlayer-cvslog] r23805 - trunk/configure

[Ivan Kalvachev]

2007/7/18, Attila Kinali <attila at kinali.ch>:
> On Tue, 17 Jul 2007 14:37:16 +0200 (CEST)
> diego <subversion at mplayerhq.hu> wrote:
>
> > Do not test for the existence of device files during configuration,
> > it fails for packagers, crosscompilers or other cases where the binaries
> > are built on a machine different from the one they are run on.
>
> Hm.. i would rather make a note in the binary packaging manual
> that distributors are encouraged to build with --enable-mga --enable-xmga
> instead of disabling this test. Most people still build mplayer
> themselves. Ie with mga_vid always enabled they will have to build
> two big drivers all the time although they will never need it
> (the mga_vid user base is quite small).
>
> Crosscompilers are a different case. Those rather tend not
> to use mga_vid as most crosscompilation is for embedded
> systems that do not use matrox graphics cards. And even
> if they would, they can easily enable it by hand during
> configure.
>
> So, i suggest to revert the mga_vid part of this commit completely.

10l cola

I complained yesterday to diego (on irc) about broken compilation on
the mga driver (something to do with latest kernel and probably missed
include).

However I'm outraged. The mga driver is now build on default on all
linux platforms, no matter if there is or not Matrox card. This should
not be so:

-If the device test is not reliable (and on properly configured system
that plugs all modules for existing devices it should be there), then
a more reliable test should be used. E.g. using lspci to find matrox
cards (it's linux tool). However I can't recall somebody complaining
from the old test.

-If distro-packages do not follow our recommendations there is no
point in forcing them. They will just disable it, anyway.

- I believe that crosscompilation also requires few more option than
usual and also extensive reading of the manual, so --enable-mga won't
be too big burden. (and as it was already pointed embedded devices
don't have mga :)

- vo_mga have higher priority because it is faster than the default
matrox xv driver. AND because vo_mga is compiled only for system with
Matrox (or by package maintainers that know what they do).

-Matrox owners are quite few (mplayer developers are not
representative group), it would be useless cruft for about 99% of
them. Makes MPlayer bigger, starting slower and outputs more alerting
messages. And slows down compilation.


I'm going to disable building of mga driver by default.

If somebody enables mga again without proper test, then I will also
enable all other obscure specific device drivers like 3dfx,tdfx,s3fb
Mind I do own Voodoo2/3 and S3VirgeDX cards and I do have enough slots
to put them all at same time


p.s.
Did I said the mga is broken... It looks like missing type definitions
(not inttypes).
--
In file included from /usr/include/linux/spinlock.h:57,
                 from /usr/include/linux/module.h:9,
                 from /usr/local/include/linux/i2c.h:41,
                 from /usr/include/linux/fb.h:5,
                 from vo_mga.c:19:
/usr/include/asm/system.h:346: error: expected declaration specifiers
or '...' before 'u8'
--
that's with headers from the 2.6.21.5 vanilla kernel.