Mac OS X backend not thread safe

[LudovicRousseau]

The Mac OS X backend uses a static variable static IOHIDManagerRef hid_mgr = 0x0;

The problem occurs with two threads are doing the sequence hid_init() [...] hid_exit()

The first thread to call hid_exit() will release hid_mgr and then the second thread will have problems.

For example I have a crash in hid_enumerate() on line:
CFSetRef device_set = IOHIDManagerCopyDevices(hid_mgr);
because hid_mgr has been reset to NULL.

My workaround is to NOT call hid_exit() so hid_mgr is not released.

A proper solution would be to add a counter of use and really release hid_mgr when no other thread is using HIDAPI anymore.

[LudovicRousseau]

This issue may be related to issues #72 and #45

[signal11]

hid_init() and hid_exit() are designed to be called at program startup and shutdown, respectively, to initialize the library. Two threads shouldn't be doing sequences of hid_init() ... hid_exit(). They only need to call hid_open*()/hid_close().

It could be beneficial to add what you're saying, but this can easily be worked around right now by using hid_init() as intended (above).

[LudovicRousseau]

In my case HIDAPI is used in a library in charge of the HID devices enumeration. This library is then used by other libraries (with no knowledge of HIDAPI). The final application can be a browser like Firefox so I cannot use hid_init() and hid_exit() at the application startup and shutdown.

A long term strategy would be to use contexts managed by hid_init() and hid_exit() and used by the other hidapi calls.

No need to fix the problem right now. I have a workaround that works for me.