Random Crash on OSX when using Thread

[tito]

On IRC, one user have few crash: the GC is triggered in a thread, and tries to deallocate a Shader. The Shader is the only instructions that are deallocated directly, it should be fixed the same way as others (texture, fbo, etc.)

Process:               Python [1255]
Path:                  /System/Library/Frameworks/Python.framework/Versions/2.7/Resources/Python.app/Contents/MacOS/Python
Identifier:            org.python.python
Version:               2.7.6 (2.7.6)
Build Info:            python-89000000000000~7
Code Type:             X86-64 (Native)
Parent Process:        pycharm [367]
Responsible:           pycharm [367]
User ID:               501

PlugIn Path:             /Applications/Kivy.app/Contents/Resources/kivy/kivy/graphics/context.so
PlugIn Identifier:       context.so
PlugIn Version:          ??? (???)

...

Thread 7 Crashed:
0   libsystem_kernel.dylib              0x00007fff8825d282 __pthread_kill + 10
1   libsystem_c.dylib                   0x00007fff8b3d1b73 abort + 129
2   org.python.python                   0x0000000105a3fa57 Py_FatalError + 49
3   libsystem_platform.dylib            0x00007fff88491f1a _sigtramp + 26
4   org.python.python                   0x0000000105a20f74 PyEval_EvalCodeEx + 1943
5   context.so                          0x0000000106bfe3a1 __pyx_f_4kivy_8graphics_7context_7Context_dealloc_shader + 41
6   shader.so                           0x0000000106123606 __pyx_tp_dealloc_4kivy_8graphics_6shader_Shader + 71
7   instructions.so                     0x000000010608d3d0 __pyx_tp_clear_4kivy_8graphics_12instructions_RenderContext + 64
8   org.python.python                   0x0000000105a52692 PyGC_Collect + 1610
9   org.python.python                   0x0000000105a52ab1 _PyObject_GC_Malloc + 213
10  org.python.python                   0x00000001059e95e5 PyType_GenericAlloc + 56
11  org.python.python                   0x00000001059baaba _PyExc_Init + 5038
12  org.python.python                   0x00000001059e9c77 _PyType_Lookup + 1254
...
88  org.python.python                   0x0000000105a26d6b PyEval_CallObjectWithKeywords + 93
89  org.python.python                   0x0000000105a53882 initthread + 1526
90  libsystem_pthread.dylib             0x00007fff922f02fc _pthread_body + 131
91  libsystem_pthread.dylib             0x00007fff922f0279 _pthread_start + 176
92  libsystem_pthread.dylib             0x00007fff922ee4b1 thread_start + 13

References: http://pastebin.com/0rwww4gH and http://pastebin.com/qwzFAp0r

[Kovak]

Looking into this further:

We do seem to 'queue' the shader deallocation just like texture/fbo.

In shader.pyx line 127

    def __dealloc__(self):
        if self.shader != -1:
            get_context().dealloc_shader_source(self.shader)

which call

context.pyx 102

    cdef void dealloc_shader_source(self, int shader):
        cdef array arr
        if shader == -1:
            return
        arr = self.lr_shadersource
        arr.append(shader)
        self.trigger_gl_dealloc()

Which clock schedule the actual cleanup. Same as texture/fbo. It seems shader was brought in line as part of fixing #955 : 1409396