The real problem is that SDL_atomic.c was built in thumb mode instead of ARM mode, which is required to use the mcr instruction on ARM platforms. Added a compiler error to catch this case so we don't generate code that does infinite recursion.
I also added a potentially better way to handle things on Linux ARM platforms, based on comments in the Chromium headers, which we can try out after 2.0.10 ships.
binarycrusader
Since changeset 358696c354a8, SDL 2.0 has been broken on Solaris when compiling with the Solaris Studio compiler (which uses the pthread implementation of SDL_AtomicLock).
Notably, it gets stuck at the MemoryBarrierRelease in SDL_GetErrBuf:
6585 # 218
6586 if (!tls_errbuf && !tls_being_created) {
6587 SDL_AtomicLock_REAL ( & tls_lock );
6588 if (!tls_errbuf) {
6589 SDL_TLSID slot;
6590 tls_being_created = SDL_TRUE;
6591 slot = SDL_TLSCreate_REAL ( );
6592 tls_being_created = SDL_FALSE;
6593 { SDL_SpinLock _tmp = 0 ; SDL_AtomicLock_REAL ( & _tmp ) ; SDL_AtomicUnlock_REAL ( & _tmp ) ; };
^^^ loops forever above
6594 tls_errbuf = slot;
6595 }
6596 SDL_AtomicUnlock_REAL ( & tls_lock );
6597 }
Running: testthread
(process id 28926)
^Cdbx: warning: Interrupt ignored but forwarded to child.
signal INT (Interrupt) in __nanosleep at 0xfe52a875
0xfe52a875: __nanosleep+0x0015: jae __nanosleep+0x23 [ 0xfe52a883, .+0xe ]
Current function is SDL_Delay_REAL
204 was_error = nanosleep(&tv, &elapsed);
(dbx) where
[1] __nanosleep(0xfeffe848, 0xfeffe850, 0xfe75a5ac, 0xfe5169d8), at 0xfe52a875
[2] nanosleep(0xfeffe848, 0xfeffe850), at 0xfe516a3b
=>[3] SDL_Delay_REAL(ms = 0), line 204 in "SDL_systimer.c"
[4] SDL_AtomicLock_REAL(lock = 0xfeffe88c), line 104 in "SDL_spinlock.c"
[5] SDL_GetErrBuf(), line 225 in "SDL_thread.c"
[6] SDL_ClearError_REAL(), line 216 in "SDL_error.c"
[7] SDL_InitSubSystem_REAL(flags = 0), line 116 in "SDL.c"
[8] SDL_Init_REAL(flags = 0), line 244 in "SDL.c"
[9] SDL_Init(a = 0), line 89 in "SDL_dynapi_procs.h"
[10] main(argc = 1, argv = 0xfeffe948), line 65 in "testthread.c"
As far as I can tell, this is because pthread_spin_trylock() always returns EBUSY for this particular lock; since it works in other places, I'm suspicious.
Different Solaris Studio compiler versions seem to make no difference.
I've verified this is broken on Linux as well if SDL_spinlock.c is modified to use the pthread implementation.
This appears to be because pthread_spin_init() and pthread_spin_destroy() are not used with the locks as required.
Andreas Ertelt
The problem in question is caused by changeset 7771 (http://hg.libsdl.org/SDL/rev/4434498bf4b9 / https://bugzilla.libsdl.org/show_bug.cgi?id=2121)
The redefinition of __inline__ (introduced by the addition of begin_code.h:128's "|| __STRICT_ANSI__") results in mingw's gcc throwing multiple
warning: always_inline function might not be inlinable [-Wattributes]
as well as a whole bunch of redefinitions of mingw internals which break linking of projects including the SDL2 headers.
