#12 pthread_spin_init used incorrectly

[Mikhail T.]

The function pthread_spin_init is invoked inside JXGrabKey.cpp with its second argument spelled out as NULL -- a pointer. That is incorrect, the second argument must be an integer and clang, for example, issues a warning.

More importantly, the NULL translates into 0, which means PTHREAD_PROCESS_PRIVATE. Whether the constant is the original NULL, or 0, or PTHREAD_PROCESS_PRIVATE, the bundled self-tests hang in a tight loop -- with the testing java-process eating up one entire CPU.

Using PTHREAD_PROCESS_SHARED (the only other legal value for the flag) allows self-tests to run to completion properly. Please, evaluate the below patch -- I do not know the API well enough...

--- JXGrabKey/C++/src/JXGrabKey.cpp 2010-02-28 10:11:49.000000000 -0500
+++ JXGrabKey/C++/src/JXGrabKey.cpp 2016-06-30 03:15:41.718599000 -0400
@@ -319,5 +320,5 @@

    XSetErrorHandler((XErrorHandler) xErrorHandler);

-   pthread_spin_init(&x_lock, NULL); // init here bcoz of the returns
+   pthread_spin_init(&x_lock, PTHREAD_PROCESS_SHARED); // init here bcoz of the returns

    doListen = true;

[Edwin Stang]

Thanks for the report. This might be due to some class sharing magic the newer JVMs include. Otherwise cannot explain why PTHREAD_PROCESS_PRIVATE should not work. Committed in Trunk, feel free to build a new release for yourself via the included ant build.xml in the misc folder.

[Mikhail T.]

Actually, trying to build (and test) it now on CentOS-6 (with OpenJDK-1.8) I observe the same hang in the test as I saw last night here. Changing the flag from PTHREAD_PROCESS_SHARED to PTHREAD_PROCESS_PRIVATE and back does not seem to make a difference:

-do-test-run:
    [junit] Testsuite: jxgrabkey.JXGrabKeyTest
    [junit] ++ listen()
    [junit] ++ registerAwtHotkey(0, 0x200, 0x46)
    [junit] registerAwtHotkey() - converted AWT mask '' (0x200) to X11 mask (0x8)
    [junit] registerAwtHotkey() - converted AWT key 'F' (0x46) to X11 keysym (0x46)
    [junit] ++ registerHotkey(0, 0x8, 0x46)
    [junit] registerHotkey() - sleeping 100 ms for listen() to be ready
    [junit] registerHotkey() - sleeping 100 ms for listen() to be ready
    [junit] registerHotkey() - sleeping 100 ms for listen() to be ready
    [junit] registerHotkey() - found in X11 mask (0x8): 'Mod1' 
    [junit] registerHotkey() - converted X11 keysym 'F' (0x46) to X11 key (0x29)
    [junit] -- registerHotkey()
    [junit] -- registerAwtHotkey()
    [junit] ++ unregisterHotkey(0)
       <........ HANGS ......>

It seems, there is a race somewhere... Once fixed, we may be able to go back to PTHREAD_PROCESS_PRIVATE as before...

Update: same problem on CentOS-7.2... Maybe, that's because I'm running test over ssh-forwarded DISPLAY (across considerable distance)?