# HG changeset patch
# User Edouard Gomez <ed.gomez@free.fr>
# Date 1224021637 -7200
# Node ID 0c4727aa17ba532172cae4aded7d16d70ea4ea53
# Parent  8e7c3f94407dc1500438237ac6bf3d484bfba742
dshow: try to improve multithreading code for RAW sample grabbing

This patch tries to address two theoritical problems:
 - The filter capturepin should not lock p_sys->lock. By chance, win32 mt
   locking is recursive by default.
    - Do not lock again, this makes things clearer for poor POSIX coders like
      me.
 - The current code does not try to push samples as fast as it receives them.
   This is caused by the arbitrary msleep call which can differ sample delivery
   with up to 10ms delay. Moreover, only a single sample at a time was processed
   either for audio or for video.
    - Use MT condition instead of sleep so no artificial delay is introduced in
      the delivery chain.
    - Process all available samples at once.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

# HG changeset patch
# User Edouard Gomez <ed.gomez@free.fr>
# Date 1224021633 -7200
# Node ID 9564c1736b76c96f37fb85476d36496b3d7955bb
# Parent  40d5c2186f7fb31e944b5b542b3366bbc420cc0e
dshow: prevent division by zero

In the DShow capture module, division by zero is possible in
multiple modulo statements when dealing with granularity values
returned by capture filters.

Change the code so that:
 - if granularity is zero, then width/height/samplingrate must be
   a perfect match with requested settings.
 - if granularity is non zero, then use the old condition tests.

This fixes real crashes reported by VLC users on different forums.

See:
http://forum.doom9.org/archive/index.php/t-134655.html
http://forum.videolan.org/viewtopic.php?f=2&t=37056&p=114065&hilit=blackmagic

Tested with both a Decklink HD Extreme card and a Decklink HD Pro card.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This fix e17, Xfce on my configuration.
Thanks very much Raster.

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Take multiple-screens into account when trying to get fullscreen controller
into center of current physical screen.

Basicly same as mplayer/libvo/x11_common.c does.

Close #2196

Signed-off-by: Rémi Denis-Courmont <rdenis@simphalempin.com>

Introduction of an abstractController, that can:
- create most of the interface controller widgets
- execute their actions

All buttons connect themselve to a mapper and a doAction( id_action )
that centralize everything. The Action could go into a mainActionner, I guess. Opinions?

This cleans a lot the signal and exchanges between MI and Controller,
between Controller and FSC. The buttons do their own cooking and connect
directly to THEMIM or to some of the few signals of the Controller
(inputExist, inputHasVideo, inputCanRecord)

This reworks most of the HACKS of Teletext Buttons and AtoB Buttons

The FSC inherit from AbstractController and not Controller, which remove the b_fscreation HACK.

There will be some regressions, I tried my best to minimize them.

The code is generic enough to be able to customize the toolbars now. HAVE FUN!

Each susbcriber has its own ring buffer already.

I don't quite understand why Qt4 would subscribe to messages all the
time rather than only when the dialog is on.

It should close #2195.