Created an attachment (id=106) [edit]
Patch for src/joystick/win32/SDL_mmjoystick.c
I am not even a Windows user, so take the following with a grain of salt:
SDL_mmjoystick.c has a function GetJoystickName which obtains the joystick
name by looking at the registry. The way it does that seems very fishy to me.
Namely, it uses the parameter "index" to construct a registry value name (BTW,
those variables used in the code are really badly named). The value of "index"
in turn equals the current value of "numdevs", as called from
SDL_SYS_JoystickInit.
I read through the MSDN docs at
<http://msdn.microsoft.com/archive/default.asp?url=/archive/en-us/dnarinput/html/msdn_extdirect.asp>,
and I believe the simple fix is to replace line 183 of said file
SYS_JoystickName[numdevs] = GetJoystickName(numdevs, joycaps.szRegKey);
by the following:
SYS_JoystickName[numdevs] = GetJoystickName(SYS_JoystickID[i],
joycaps.szRegKey);
However, that is only *hiding* the real issue. Problem is, the list of
joysticks as returned by windows may contains "gaps", and the code deals
incorrectly with that. Namely those gaps occur if joysticks are
removed/(re)added, as the reporter observed.
The attached patch fixes this and another (off-by-one) issue in the code. But
since I have no Windows machine, I can't even test-compile it, so use with
caution.
> BTW, when setting up parallel make, I usually use # cpus + 1, so a compile is
> running while disk access is going for another.
[From Ryan]
My experience is that this works well on Linux, but is actually slower on
PowerPC Mac OS X...not sure if that's an architecture issue or a scheduler
issue, though, and haven't tried it on Intel Mac OS X.
- A change to define CXX in fatbuild, which comforts the configure script a little, even if we don't use C++ anywhere.
- Some code to see how many CPU cores exist and parallelize make across them.
- CFLAGS that apply to both archs are specified seperately (-O3, -pipe, etc)
- -fvisibility=hidden for the gcc4 builds
- a "clean", "clean-ppc" and "clean-x86" command
- Fix to SDL_config_macosx.h for the HAVE_ALLOCA_H thing.
Now builds on an Intel Mac.
From: Julien Lecomte
Subject: [SDL] sdl-config
I've noticed that `sdl-config' usage reports that --static-libs is
available even when it's not.
I've modified the 'sdl-config.in' file so that it no longer reports
'--static-libs' if SDL was compiled with --disable-static.
I didn't change --disable-shared behavior.
The clipping is done at a higher level, and the low level functions are
passed clipped rectangles. Drivers which don't support source clipping
have not been changed, so the image will be squished instead of clipped,
but at least they will no longer crash when the destination rect was out
of bounds.
On servers with the composite extension enabled, visuals with 32-bit depth
have an alpha mask. This is pretty neat, but SDL needs a bit more work to
handle these properly, so for now, we'll just use 24 bit depth visuals.
From: Mike Frysinger
Subject: [SDL] [patch] touchup configure --help output
the current output includes a lot of "default=yes" stuff in it ... except that
it lacks the [] around it
patch attached to fix the situation
From Christian Walther:
When writing my patch for #12, I ended up doing all sorts of changes to the way
application/window activating/deactivating is handled in the Quartz backend,
resulting in the attached patch. It does make the code a bit cleaner IMHO, but
as it might be regarded as a case of "if it ain't broken, don't fix it" I'd
like to hear other people's opinion about it. Please shout if some change
strikes you as unnecessary or wrong, and I'll explain the reasons behind it. As
far as I tested it, it does not introduce any new bugs, but I may well have
missed some.
- The most fundamental change (that triggered most of the others) is irrelevant
for the usual single-window SDL applications, it only affects the people who
are crazy enough to display other Cocoa windows alongside the SDL window (I'm
actually doing this currently, although the additional window only displays
debugging info and won't be present in the final product): Before, some things
were done on the application becoming active, some on the window becoming key,
and some on the window becoming main. Conceptually, all these actions belong to
the window becoming key, so that's what I implemented. However, since in a
single-window application these three events always happen together, the
previous implementation "ain't broken".
- This slightly changed the meaning of the SDL_APPMOUSEFOCUS flag from
SDL_GetAppState(): Before, it meant "window is main and mouse is inside window
(or mode is fullscreen)". Now, it means "window is key and mouse is inside
window (or mode is fullscreen)". It makes more sense to me that way. (See
http://developer.apple.com/documentation/Cocoa/Conceptual/WinPanel/Concepts/ChangingMainKeyWindow.html
for a discussion of what key and main windows are.) The other two flags are
unchanged: SDL_APPACTIVE = application is not hidden and window is not
minimized, SDL_APPINPUTFOCUS = window is key (or mode is fullscreen).
- As a side effect, the reorganization fixes the following two issues (and
maybe others) (but they could also be fixed in less invasive ways):
* A regression that was introduced in revision 1.42 of SDL_QuartzVideo.m
(http://libsdl.org/cgi/cvsweb.cgi/SDL12/src/video/quartz/SDL_QuartzVideo.m.diff?r1=1.41&r2=1.42)
(from half-desirable to undesirable behavior):
Situation: While in windowed mode, hide the cursor using
SDL_ShowCursor(SDL_DISABLE), move the mouse outside of the window so that the
cursor becomes visible again, and SDL_SetVideoMode() to a fullscreen mode.
What happened before revision 1.42: The cursor is visible, but becomes
invisible as soon as the mouse is moved (half-desirable).
What happens in revision 1.42 and after (including current CVS): The cursor is
visible and stays visible (undesirable).
What happens after my patch: The cursor is invisible from the beginning
(desirable).
* When the cursor is hidden and grabbed, switch away from the application using
cmd-tab (which ungrabs and makes the cursor visible), move the cursor outside
of the SDL window, then cmd-tab back to the application. In 1.2.8 and in the
current CVS, the cursor is re-grabbed, but it stays visible (immovable in the
middle of the window). With my patch, the cursor is correctly re-grabbed and
hidden. (For some reason, it still doesn't work correctly if you switch back to
the application using the dock instead of cmd-tab. I haven't been able to
figure out why. I can step over [NSCursor hide] being called in the debugger,
but it seems to have no effect.)
- The patch includes my patch for #12 (it was easier to obtain using cvs diff
that way). If you apply both of them, you will end up with 6 duplicate lines in
SDL_QuartzEvents.m.
The SDL platform header shouldn't dictate the OS target API for other
applications. The build system should define the target API for building
SDL internally.
SDL_SemValue() always returns 0. That's because the underlying POSIX
sem_getvalue() is implemented as a noop in Mac OS X. Apart from that,
semaphores do work properly (at least according to test/testsem), so I'm not
sure if this is worth fixing at all.
[I'm fixing this for the public headers, but I'm not going to bother for the SDL library code (yet)]
To clarify: Normaly, GCC (or, to be precise, the preprocessor) will ignore
this, and compile the code happily. However, one can specify -Wundef to get a
warning about this.
One can probably argue whether to consider this a bug or not; but I think that
(a) from a semantic point of view, using "#if FOO" when FOO is not defined is
strange, and (b) since it is possible to trigger a warning about this, and a
trivial fix exists, it should be corrected.
I can think of two alternative patches, BTW:
1) Simply use #define HAVE_FOO 0, instead of not defining HAVE_FOO at all
2) Change
#if HAVE_FOO
to
#if HAVE_FOO+0
which always does the right thing.
But I think I still prefer the attached patch :-).
The attached patch is mostly cosmetic: Currently, we always add both -framework
OpenGL and -framework AGL for Mac OS X command line builds.
However, the former is only used for the Cocoa video driver, and the latter
only for the Carbon video driver (as far as I can tell, at least). Hence the
attached patch modifies configure.in so that each only gets added to the list
of frameworks if the corresponding video driver is enabled.
The change made to sdl-config.in on March 18 (rev 11, see
http://www.libsdl.org/cgi/cvsweb.cgi/SDL12/sdl-config.in), broke static linking
under Mac OS X for unix style build systems. In other words, "sdl-config
--static-libs" no longer produces the correct output, as it is missing the
required -framework switches.
The attached patch fixes this.
[I opted to go for a warning, since I often tweak configure.in, but don't want to rebuild the entire project]
One thing that was lost in the switch from automake to the new build system is
that there is now no rule to build configure from configure.in.
IMHO, if configure.in gets changed, then at the very least, the build system
should print out a warning (better, again IMHO, an error) about this fact.
Else, you easily forget about this when modifying configure.in.
The proper name of Apple's operating system is "Mac OS X" not "MacOS X", as can
bee seen in many places, for example http://www.apple.com/macosx/). This
contrasts the naming of the old operating system, which was called "MacOS" and
today is often refered to as "MacOS Classic".
The attached patches fixes the misuse of the name "MacOS X" in both the SDL12
and sdlweb CVS modules.
The current definition of int32_t, uint32_t, uintptr_t
doesn't match what the Mac system headers already have...
Since it's an old 16/32 bit platform, the system headers
expect the types to use "long" and not "int" like they do.
