duplication
 . YUV transformations now use the new module API ; now only the vout
   and interface still depend on the old plugin API
 . changed TestMMX to TestCPU because we will be able to test for other
   specific extensions (3DNow, SSE) for the forthcoming IDCT we'll
   borrow to mpeg2dec (as usual :P)
 . don't show "hiding module" messages anymore except in debug mode
 . swapped C IDCT and Classic IDCT scores since the classic one seems
   to be faster ; anyway you should use the MMX one
 . fixed a Makefile bug under BeOS
 . tried to fix the cpuid code so that it compiles under BeOS (Polux, can
   you test it when you have time ?)

* Fixed a bug in the DTS/PTS ordering.
Our pts are _at last_ accurate.

* Fixed --width and --height support in SDL

   OK, so I foired input.h once again. But this one works. For real.

 . fixed the bug-that-made-the-vlc-segfault-on-exit, which means that
   more than ever, � le client ne segfaulte pas �.

 . there was still a problem upon quitting: sometimes the null packet
   wasn't big enough to reach a decoder breakpoint, so I now create 10
   of them (since I can't make it bigger) to be sure. Meuuh, what do
   you think ?

 . fixed the Makefile to spare a few variables here and there.

 . reduced module hide delay, removed loading of the audio output
   plugins since they are now modules.

 . changed a few intf_Msg to intf_DbgMsg, and added "vout:", "intf:",
   and so on in a few other messages.

 . removed unused includes in the idct modules.

initialized. It remains some segfaults but it globally works.
. DVD mode is enabled with option --dvd. For the moment, it only read
the first vts and find the movie if it is at the beginning of the title
set.

Coming soon:
. udf filesystem management to access over 2Gb zones.
. detection of the offset to the movie inside a vts when it is not at
the beginning.

Is there a chance that using fopen/fseek function instead of open/lseek
system calls solve the 2Gb issue ?

. Parsing of ifo file almost completed.
. Still does not work well.

 . second part of the commit

 . the IDCT functions are now located in modules : the classic IDCT,
   our optimized IDCT, and MMX IDCT.
 . cosmetic changes for modules (pf_* instead of p_*)
 . fixes to the BeOS and dummy modules

   WARNING: a new directory has been added, which means you have to
   use `cvs update -dP' to update ; files have been removed , which means
   you either have to run `make distclean' or remove the .deps directory,
   and the configure script has changed, which means you have to run it
   again. Don't complain before having checked this :)

   Note about the IDCT modules: there is no way to choose the one we
   use, but you may remove the module in lib/ to prevent it from being
   loaded. The preference order is : MMX, Optimized, Classic.

INPUT_PTS_DELAY down to 200 ms ;
video heap down from 10 to 5 pictures ;
decoder fifo size from 1023 to 511 PES ;
* Fixed various bugs in 32 bit-versions of input_ext-dec.h ;
* Fixed a bug in GetChunk() ;
* Renamed GetByte, GetWord, ShowWord to _GetByte, _GetWord, _ShowWord ;
* Moved decoder_fifo-specific code from programs.c to dec.c ;
* Fixed bugs in program.c that prevented vlc to close all decoders ;
* Gave sam a lesson for the use of the bitstream in spu_decoder.c :ppp ;
* Made the video parser unlock the reference pictures before quitting
(still one left, yaknow why ?)

   the last time. henri, that'll be one binouze.

    probably need a few fixes to work or even compile.

Done :
- ported the alsa plugin to the news vlc module API
- cosmetic changes

To do :
- check if it works with the latest alsa release
- add features ... prepare for 4 or more speakers

  . moved the MPEG audio decoder to the new bitstream

    I wasn't sure whether the bits counting method was effective to
  skip ancillary data at the end of a frame, but I kept it for
  safety. There is still a lot to optimize in the bit counting, like
  unrolling the first iteration of some loops, but I don't know if
  it would be worth it. The other solution would have been to look for
  a startcode after eache frame, without caring about the ancillary bits.

  . cleaning in the SPU decoder.

 . moved swab32 from input_ext-dec.h to common.h. We probably won't need
   it anymore really soon, since ntohl() and htonl() seem to be properly
   optimized when gcc is passed the right optimization flags.

 . fixed GetBits32 and RemoveBits32. I know the #ifdefs are useless since
   we only support u32 words, but it's a safe reminder. Comments appreciated
   on this fix since I may have b0rked something -- it runs well here though.

 . fixed a missing #ifdef STATS
    *grin*

* Move input_ext-dec.h to input.h ;
* New input_NullPacket function ;
* Fixed init bugs in input and vpar ;

* Fixed a purify bug in input_ps.c ;
* Added -mcpu=pentiumpro in the Makefile.

 . complete commenting of modules_core.h and small modifications
 . separated MODULE_CONFIG_START and ADD_WINDOW (Henri, beware)
 . fixed a very, very old bug that caused namespace collision between
   plugins (Henri, beware: call your aout_getplugin function like this,
   not alsa_aout_getplugin)
 . removed the Debian specific clean rules from the Makefile, since they
   are already handled by the debian/rules script.
 . tidied a few compilation rules (-lfoo should always be at the end)

 . just removed a forgotten debug message in the setpalette code

.Tried to add it to stable, failed, probably because of the initialisation codes which are differents...

 . fixed a warning under FreeBSD (dlerror() is a const char*, not a char*).
 . made configure script look for the bswap assembly instruction (using
   AC_TRY_COMPILE makes sure the compiler will really accept it).

    from RGB masks);
  - cleaning of the fullscreen switch, still sucks but the surface is not
    reallocated and that's good (thanks to the magic function
    SDL_WM_ToggleFullScreen provided by sdl library :).

 . Added a missing mutex_unlock in video_output.c
 . Moved a mutex_destroy after the thread cancellation to prevent
   possible problems
 . Added a missing #ifdef in intf_msg.c which prevented compilation
   with --enable-debug
 . Fixed a bug in the input that prevented to quit properly in certain
   cases (some ES were not deleted because their index was skipped when
   the former one was deleted, well Meuuh should know what I mean)

  There is still a segfault when quitting, most presumably in the
 audio output. I couldn't find what causes it yet.

   removed whenever one does a cvs update -dP
 . fixed debian/.cvsignore because of a new version of debhelper

 . made the audio_output API a bit simpler.

 . got rid of modules_config.h.

 . fixed `make install' rule.

 . fixed warnings in the MMX YUV compilation.

 . probably fixed a bug in the input : pp_foo structures were free()d at
   the end, but this was useless since the last call to realloc() is
   equivalent to free(), and two consecutive calls to free() aren't a
   goo idea.

 . we check that the version number match and that we don't already have
   a module by that name when loading a new module.

 . all public module_* functions now properly lock the module bank.

 . the audio_output now uses the new module API ; EsounD and DSP modules
   have been ported, which should be enough for Henri to port the ALSA one.


   The new plugin API is now much simpler ; it's now just a matter of
calling module_Need( p_main->p_module_bank, MODULE_CAPABILITY_FOO, p_data );
and if successful we get the best module. Capabilities can be ORed, so that
one can ask for a plugin that does VOUT _and_ INTF actions for instance.

   It's not complete yet though -- by making it generic I had to do a few
ugly casts here and there, which I am going to fix ASAP. Also, command line
selection of a plugin does not work yet.

   The switch to the new plugin API has probably broken the BeOS audio
output ; we can either wait until the whole vlc has switched to the new
plugins, or create a separate beos_sound.so that conforms to the new API.

vais me coucher.

- Order : if a then b are initialized, release b then a ;
- Typos ;
- Cosmetic changes.

- Added vlc_mutex_destroy and vlc_cond_destroy function, for pthreads.
- Used them before quitting, every lock and cond is destroyed
- Checked the return value of malloc and realloc in input_programs
- Cosmetic changes

TODO: add vlc_*_destroy for beos and C_thread

 . should compile & run on Solaris with ./configure --disable-dsp
    (tested on puma)
 . removed most SYS_SOLARIS #ifdef's in favour of more generic tests
 . updated example ./configure line in INSTALL file
 . now we only load modules that end with ".so"
 . null module is compiled by default