Hans de Goede 2009-02-13 01:10:52 PST
Since the new "glitch free" version of pulseaudio (used in Fedora 10 amongst
others), the sound of SDL using apps (like a simple playmus call) has been
crackling.
While looking in to fixing this I noticed that the current pulseaudio code in
SDL uses pa_simple. However pa_simple uses a thread to pump pulseaudio events
and ipc, given that SDL already has its own thread for audio handling this is
clearly suboptimal, leading to unnecessary context switching IPC, etc. Also
pa_simple does not allow one to implement the WaitAudio() callback for SDL
audiodrivers properly.
Given that my work is mostly a rewrite (although some original pieces remain)
I'm attaching the new .c and .h file, as that is easier to review then the huge
diff.
Let me know if you also want the diff.
This new version has the following features:
-no longer use an additional thread next to the SDL sound thread
-do not crackle with glitch free audio
-when used with a newer pulse, which does glitch free audio, the total latency
is
the same as with the alsa driver
-proper WaitAudio() implementation, saving another mixlen worth of latency
-adds a WaitDone() implementation
This patch has been written in consultancy with Lennart Poetering (the
pulseaudio author) and has been reviewed by him for correct use of the pa API.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403856
Scott McCreary 2008-08-21 10:48:14 PDT
This patch adds support for Haiku.
http://ports.haiku-files.org/browser/haikuports/trunk/media-libs/libsdl/SDL-1.2.13-haiku.diff
Haiku is an open-source recreation of BeOS. It has better POSIX compliance
than beOS did, and other improved features, which in some cases causes us to
have to "undo" previous BeOS workarounds.
Here's our port log entry for it, showing the steps to force the changes into
configure and Makefile:
http://ports.haiku-files.org/wiki/media-libs/libsdl/1.2.13/1
Note that this was only tried on 1.2.13 stable so far.
Haiku is using a newer config.guess / config.sub that doesn't yet seem to be in
the released libtool, so we are having to copy it in for now.
http://haiku-files.org/files/optional-packages/
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403854
Pierre Phaneuf 2008-05-01 10:29:16 PDT
SDL_VIDEO_DRIVER_X11 is not defined in SDL_config_macosx.h
There's a number of SDL_VIDEO_DRIVER_xxx defined in SDL_config_macosx.h, but
not that one, even though it is actually present.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403849
Wade Berrier 2008-05-28 11:09:42 PDT
configure.in will enable dynamic loading by default for modules, but some of
the checks only check /lib, /usr/lib, /usr/local/lib, etc...
It needs to also account for systems using /lib64, /usr/lib64, and
/usr/local/lib64.
For example, openSUSE x86_64 distro uses these style paths. Otherwise, i586
gets dynamically loaded, but x86_64 doesn't.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403848
Please merge this patch for the PA driver in SDL.
http://0pointer.de/public/sdl-pulse-rework.patch
This patch:
- fixes buffering (i.e. reduces number of "fragments" to 2, doesn't defer
filling up of the buffer until the entire buffer ran completely empty.)
- drops $PASERVER and $PADEVICE env var support, since this is a duplication of
$PULSE_SERVER and $PULSE_SINK which the PA libs honor anyway.
This fixes the sound issues in all games I tested.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403846
SDL_SemWaitTimeout in src/thread/generic/SDL_syssem.c line 179 (SVN trunk):
--sem->count;
should be
if (retval == 0) {
--sem->count;
}
Without this, sem->count will underflow on timeout effectively breaking the
semaphore. It appears that the implementation has been wrong since the initial
revision.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403843
this patch by Diego Pettenò <flameeyes@gentoo.org> for SDL-1.2:
The attached patch applies over latest ~arch SDL version, and allows to use the
xinerama support to launch an application full screened on the head "0".
The SDL_VIDEO_FULLSCREEN_HEAD environment variable sets the head on which the
full screen will be displayed, but if you set it to 0, the code simply ignores
it as unset. My patch changes the unset value to -1, so that the 0 value can be
used correctly. Without this, trying to get fullscreen on head 0 would get the
same result than not having xinerama enabled at all.
http://sources.gentoo.org/media-libs/libsdl/files/libsdl-1.2.11-xinerama-head-0.patch
SDL-1.3 doesnt seem to have any code like this, but it's still good for SDL-1.2
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403841
Now we use the software path for YUV, and CoreGraphics for 2D stuff.
There are several other 10.6 fixes in here, too...now we can build a 64-bit
SDL for Snow Leopard!
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403838
It's an OS 9 API that was exposed via Carbon, but is gone in the 10.6 SDK.
OS X uses CoreAudio. We only had it in here for debugging purposes anyhow,
as SDL would never use it unless forced to with an environment variable.
It remains in SDL for OS 9, of course.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403792
Fixed freeze when trying to memset with 0 length
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403757
From: Stefan Klug
Subject: Re: [SDL] SVN doesn't compile for wince
the patch applied for Revision 4483 was seemingly not checked for side
effects.
It broke the WinCE build.
The attached patch should fix these problems. I'm not using SDL 1.2 on
CE anymore, and therefore haven't tested the patch... but at least it
compiles ;-)
Regards Stefan
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403602
Comment #3 From esigra@gmail.com 2008-04-09 11:46:46 (-) [reply] -------
This bug also needs to be fixed for SDL_GetRGB and SDL_GetRGBA (those should
not change the format of the surface either). Until this is fixed, I still need
to const_cast the format parameter in calls to those functions.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403581
Description From maf6@cornell.edu 2009-01-10 14:07:52 (-) [reply]
Assuming the SDL project doesn't just want to pick these up from dinput8.lib or
whatever instead, the definitions in SDL_dx5video.c should be:
const DIDATAFORMAT c_dfDIKeyboard = { sizeof(DIDATAFORMAT),
sizeof(DIOBJECTDATAFORMAT), 0x00000002, 256, 256, KBD_fmt };
const DIDATAFORMAT c_dfDIMouse = { sizeof(DIDATAFORMAT),
sizeof(DIOBJECTDATAFORMAT), 0x00000002, 16, 7, PTR_fmt };
const DIDATAFORMAT c_dfDIJoystick = { sizeof(DIDATAFORMAT),
sizeof(DIOBJECTDATAFORMAT), 0x00000001, 80, 44, JOY_fmt };
This way the call to IDirectInputDevice2_SetDataFormat in SDL_dx5events.c will
succeed on x64.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403580
Description From Tim Angus 2008-08-31 03:38:36 (-) [reply]
When pressing "system keys" e.g. ALT+F4 etc., the default for the windib driver
is to ignore them completely, whereas with the directx driver they're handled
(badly as it turns out).
As the windib driver is the default, I assume its behaviour is considered
correct, so the following patch makes the directx driver the same in this
regard.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403579
Description From Tim Angus 2008-08-30 12:23:56 (-) [reply]
As we all know SDL 1.2 doesn't handle dead keys well since one key press
potentially equals two (or more) characters. For example, on many layouts,
keying <backquote>,<space> results in <no character>,<backquote><space>. Since
the unicode member of the SDL_keysym struct only has room for one character,
only one can be returned.
On Linux, the first character is returned. On Windows however, unless the exact
number of characters generated by the keypress is 1, nothing is returned. The
following patch addresses this inconsistency.
Updated patch which includes a further fix to the handling of the numpad when
numlock is on. This further fix is courtesy Amanieu d'Antras.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403578
Comment #22 From Tim Angus 2009-04-02 08:45:52 (-) [reply] -------
First of all, thanks for committing this. Unfortunately it seems the patch has
only partially applied to wincommon/SDL_sysevents.c and currently a clean SDL
1.2 checkout doesn't build. The new patch here
(http://bugzilla.libsdl.org/attachment.cgi?id=316) fixes this.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403577
Comment #1 From Simon Howard 2009-03-20 16:50:56
Hi,
I'm the author of Chocolate Doom, one of the other source ports that James
mentioned. This is a patch against the current SVN version of SDL 1.2 that
fixes the bug. It has been tested and hopefully should be obviously correct
from examining the changes. I'll give a brief explanation.
When the palette is set with SDL_SetPalette, the IDirectDrawPalette_SetEntries
DirectX function is invoked. However, when this happens, a WM_PALETTECHANGED
message is sent to the window.
A WM_PALETTECHANGED message can also be received if the palette is changed for
some other reason, like if the system palette is changed. Therefore, the
palette change handler (DX5_PaletteChanged) has code to deal with this case.
It distinguishes "expected" palette changes (set with SDL_SetPalette) from
"unexpected" palette changes using the colorchange_expected variable, which is
set before calling IDirectDrawPalette_SetEntries. However, the code to set
this variable is missing in the fullscreen code path. By setting this
variable, the palette change is handled properly and the freezes go away.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403576
From Tim Angus 2008-08-12 11:18:06
I'm one of the maintainers of ioquake3.org, an updated version of the
Quake 3 engine. Relatively recently, we moved ioq3 to use SDL as a
replacement for 95% of the platform specific code that was there. On the
whole it's doing a great job but unfortunately since the move we've been
getting complaints about the quality of the mouse input on the Windows
platform to the point where for many the game is unplayable. Put in
other terms, the current stable SDL 1.2 is basically not fit for purpose
if you need high quality mouse input as you do in a first person shooter.
Over the weekend I decided to pull my finger out and actually figure out
what's going on. There are basically two major problems. Firstly, when
using the "windib" driver, mouse input is gathered via the WM_MOUSEMOVE
message. Googling for this indicates that often this is known to result
in "spurious" and/or "missing" mouse movement events; this is the
primary cause of the poor mouse input. The second problem is that the
"directx" driver does not work at all in combination with OpenGL meaning
that you can't use DirectInput if your application also uses OpenGL. In
other words you're locked into using the "windib" driver and its poor
mouse input.
In order to address these problems I've done the following:
* Remove WM_MOUSEMOVE based motion event generation and replace with
calls to GetCursorPos which seems much more reliable. In order to
achieve this I've moved mouse motion out into a separate function that
is called once per DIB_PumpEvents.
* Remove the restriction on the "directx" driver being inoperable in
combination with OpenGL. There is a bug for this issues that I've
hijacked to a certain extent
(http://bugzilla.libsdl.org/show_bug.cgi?id=265). I'm the first to admit
I don't really understand why this restriction is there in the first
place. The commit message for the bug fix that introduced this
restriction (r581) isn't very elaborate and I couldn't see any other bug
tracking the issue. If anyone has more information on the bug that was
avoided by r581 it would be helpful as I/someone could then look into
addressing the problem without disabling the "directx" driver.
* I've also removed the restriction on not being allowed to use
DirectInput in windowed mode. I couldn't see any reason for this, at
least not from our perspective. I have my suspicions that it'll be
something like matching up the cursor with the mouse coordinates...
* I bumped up the DirectInput API used to version 7 in order to get
access to mouse buttons 4-7. I've had to inject a little bit of the DX7
headers into SDL there as the MinGW ones aren't up to date in this respect.
--HG--
branch : SDL-1.2
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/branches/SDL-1.2%403572
