* SDL_timer_running wasn't always updated correctly.
* Fixed occasional crash in SDL_SetTimer() when clearing threaded timers
* It was possible to get both the timer thread and event thread running
* Other misc. cleanup
From: Jon Daniel
Subject: [SDL] KeyRepeat fix
If you switch keys very fast and hold the last one it
doesn't get repeated because the key release of the first key
occured after the key press of the second key.
Unfortunatly the key release of the first key sets
SDL_KeyRepeat.timestamp = 0; causing the the second key not to repeat.
This should be always reproducable. I've only checked this on x86-64
X11.
To fix this I just added another condition to make sure the sym of the
released key matches SDL_KeyRepeat.key.keysym.sym.
From: Hayashi Naoyuki
Subject: SDL patch for Tru64 UNIX 4.0X
SDL-1.2.8 support only Tru64 5.X.
This patch is for Tru64 UNIX 4.X.(tested on Tru64 UNIX 4.0G and 5.1B)
SDL-1.2.8/configure.in:
ld doesn't accept -pthread option.
cc -pthread has same effect as -D_REENTRANT when compiling,
and has same effect as -lpthread -lexc when linking.
SDL-1.2.8/include/begin_code.h:
Old Compaq C Compiler accept not inline but __inline.
SDL-1.2.8/src/audio/SDL_mixer_MMX.c:
SDL-1.2.8/src/audio/SDL_mixer_MMX.h:
Old Compaq C Compiler doesn't accept //.
SDL-1.2.8/src/cdrom/osf/SDL_syscdrom.c:
When becoming Tru64 v5.0 from Tru64 v4.0,
the arrangement of the cd-rom device was changed.
From: Mike Frysinger
Subject: [SDL] minor fix for gcc-2.x building
in src/video/directfb/SDL_DirectFB_video.c, some variables are not declared at
the beginning of scope ... newer gcc's will accept this, but gcc-2.x does not
find attached a simple patch by Matt Taylor to resolve this
http://bugs.gentoo.org/show_bug.cgi?id=75392
From: Marcin Konicki
Subject: Re: [SDL] SDL 1.2.8 Prerelease
I'm sending small fix for BeOS, which prevents filling up SDL's message
queue too fast.
Without it, SDL receives "key down" messages from BeOS code, for each key
repeat (BeOS handles key repeats itself, and application can check if
received "key down" message from BeOS is first time key down, or if it's
repeat, and which repeat it is). Since there is no way for "sdl driver" to
turn off "default" SDL's key-repeat mechanism, they were working both at
the same time (and queue could be filled up very fast).
So this patch removes handling "key down" message from BeOS if it's
key_repeat "type".
From: Jonathan Atkins
Subject: [SDL] SDL_SaveBMP width bugfix
this fixes the pitch versus width difference that can happen
(especially for 8bit and 24bit (with the exact RGBAmasks) surfaces)
when you use SDL_SaveBMP. The problem was the pitch was used
instead of the width, which in some cases is much wider than the
screen area you really want to save...making for ugly crud on the
saved image borders.
This code has been tested with & without pitch overhangs...and
with the right masks for 24 bit surfaces.
I tested 8,15,16,24,32-0RGB,32-RGBA(with no SDL_SRCALPHA flag).
From: Stephane Marchesin
Subject: Re: [SDL] Problem compiling SDL 1.2.7
- there is a bug that was introduced in the kernel headers for 2.6.9
which is fixed in 2.6.10. This bug *will* byte when compiling the cdrom
subsystem. A patch that works around this bug is attached. Note that
users affected are not those running 2.6.9, but those using the 2.6.9
kernel headers for their system (i.e. whose libc is built against 2.6.9
headers).
From: "Martin Bickel"
Subject: [SDL] Patch: fixing uninitilized palette
while running Valgrind over my application I found the following
problem in SDL:
The function MapNto1 allocates SDL_Color colors[256] but does not
initialize it.
SDL_DitherColors is then called which initialized the r, g and b
component, but not the 'unused' component of each color.
When Map1to1 is called from MapNto1, it runs a memcmp on the colors,
which also evaluates the unused component and therefor returns
differences much more often than necessary.
So the 'unused' component of SDL_Color should be initialized. This
patch does this by calling memset for the whole array in MapNto1 .
From: Eric Wing
Subject: OS X Mouse inversion problem fix (again)
Here's yet another patch for the OS X mouse inversion
problem. This should fix the problem once and for all.
I know I've said this before, but *This time for
sure!* :)
If you recall, my last patch broke the non-OpenGL
windowed code and caused the inversion to occur there
instead. Max submitted a patch that partially reverted
the changes back which included the os version hack
which is currently the most recent CVS.
Aaron Sullivan identified and reported to the mailing
list the other day, that the last partial regression
of the code broke OS X 10.2. Looking over the results,
I'm thinking that I was slightly more successful than
I thought at unifying the code. I think I was trying
to unify the code base for OpenGL and non-OpenGL
windowed modes for all versions of the OS. It looks
like I failed at at unifying the OpenGL and non-OpenGL
code, but I did succeed at unifying the OS versions.
Thus, we no longer need the hack for the OS version
checks. The partial regression still included an OS
check which is what broke things for < 10.3.
Attached is the patch for SDL_QuartzWM.m. It basically
is a half-line change that removes one of the two
checks that decides if the mouse coordinates need to
be inverted, i.e:
if (system_version >= 0x1030 &&
(SDL_VideoSurface->flags & SDL_OPENGL) )
becomes this:
if(SDL_VideoSurface->flags & SDL_OPENGL)
With Aaron's outstanding help, we have collectively
tested:
windowed OpenGL
windowed non-OpenGL
fullscreen OpenGL
fullscreen non-OpenGL
under OS X 10.2 (Jaguar), 10.3 (Panther), and 10.4
(Tiger).
We don't have access to 10.0 or 10.1, but since the
original problem didn't materialize until 10.3, I'm
hopeful that testing 10.2 is sufficient. And now that
the code is uniform, I'm also hoping we'll be safe
moving forward to deal with future revisions of the OS
with this issue.
