nullpointerexception in GC

[ddyer0]

I was debugging under xcode, and got a nullpointer exception in the GC thread at the indicated point.

JAVA_BOOLEAN com_codename1_impl_ios_IOSNative_isSocketConnected___long(CN1_THREAD_STATE_MULTI_ARG JAVA_OBJECT instanceObject, JAVA_LONG socket) {
POOL_BEGIN();
SocketImpl* impl = (BRIDGE_CAST SocketImpl*)((void *)socket);
JAVA_BOOLEAN b = [impl isConnected]; // HERE
POOL_END();
return b;
}

stack trace
Thread 3

#0 0x00000001136091a2 in virtual_java_lang_Long_longValue___R_long at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/java_lang_Long.m:506
#1 0x00000001114307d6 in com_codename1_impl_ios_IOSImplementation_isSocketConnected___java_lang_Object_R_boolean at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_impl_ios_IOSImplementation.m:14456
#2 0x0000000110f75a02 in virtual_com_codename1_impl_CodenameOneImplementation_isSocketConnected___java_lang_Object_R_boolean at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_impl_CodenameOneImplementation.m:16156
#3 0x0000000113d554fe in com_codename1_io_Socket_SocketInputStream_close__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_io_Socket_SocketInputStream.m:116
#4 0x0000000113d578c7 in virtual_com_codename1_io_Socket_SocketInputStream_close__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_io_Socket_SocketInputStream.m:462
#5 0x0000000113d54ea3 in com_codename1_io_Socket_SocketInputStream_finalize__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_io_Socket_SocketInputStream.m:398
#6 0x0000000113d54a5d in FINALIZER_com_codename1_io_Socket_SocketInputStream at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/com_codename1_io_Socket_SocketInputStream.m:59
#7 0x000000010f699258 in freeAndFinalize at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/cn1_globals.m:430
#8 0x000000010f699d38 in codenameOneGCSweep at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/cn1_globals.m:830
#9 0x00000001100fc817 in java_lang_System_gcMarkSweep at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/nativeMethods.m:1116
#10 0x0000000111449d67 in java_lang_System_access$200__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/java_lang_System.m:342
#11 0x000000010f794556 in java_lang_System_1_run__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/java_lang_System_1.m:120
#12 0x0000000110add2dc in virtual_java_lang_Runnable_run__ at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/java_lang_Runnable.m:60
#13 0x000000011030528d in java_lang_Thread_runImpl___long at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/java_lang_Thread.m:248
#14 0x00000001100fcde0 in threadRunner at /Users/admin/Desktop/boardspace-7.2/dist/Develop-src/nativeMethods.m:1337

All of the implementation seems to assume that "socket" has been allocated, but that is clearly
not the case.

[shai-almog]

Might be related to my work on #3726

[ddyer0]

i agree. if the finalizer is new that would explain how it was unnoticed. I'm still tracking the issue, but I haven't seen
any directly bad behavior that corresponds to this other than in the debugger.

[ddyer0]

The issue I’m tracking is compatible with the hypotheses that this npe leaves the gc disabled, and subsequently any process that becomes blocked for gc is effectively dead.

Additional evidence: The error I'm tracking first occurred in my logs in a build I made on Sep 15

I hacked my use of sockets to prevent my SocketInputStream and SocketOutputStream instances from being
garbage collected, and that seems to have fixed it.

I don't have a good model how this could have wedged the garbage collector without causing a more
immediate problem - do you ? - It seem to me that there should have been some immediate effect rather
than leave it to victimize some random process sometime in the future in this mysterious way.