Bug 6135 – Thread/GC interaction bug on OS X

Consider the following program: --- import core.memory; import core.thread; void doNothing() {} void main() { foreach (i; 0 .. 10000) { auto t = new Thread(&doNothing); t.start(); GC.collect(); } } --- With latest DMD (9aae5c4) and druntime (00ce4ab) on OS X, this triggers the "Unable to load thread state" clause at core/thread.d:2280 during a GC collection (which in turn triggers an onOutOfMemoryError() on allocating the exception object because it happens during a GC runs, which accounts for #0-#5 in the backtrace below): --- #0 0x00012787 in _d_throwc () #1 0x00009686 in onOutOfMemoryError () #2 0x0000e8b9 in D2gc3gcx2GC12mallocNoSyncMFkkPkZPv () #3 0x0000e855 in D2gc3gcx2GC6mallocMFkkPkZPv () #4 0x0000dc58 in gc_malloc () #5 0x000134bb in _d_newclass () #6 0x0000b558 in D4core6thread17thread_suspendAllUZv7suspendMFC4core6thread6ThreadZv () #7 0x0000b447 in thread_suspendAll () #8 0x00010b75 in D2gc3gcx3Gcx11fullcollectMFPvZk () #9 0x00010b2d in D2gc3gcx3Gcx16fullcollectshellMFZk () #10 0x0000faf7 in D2gc3gcx2GC11fullCollectMFZk () #11 0x0000da59 in gc_collect () #12 0x000096bb in D4core6memory2GC7collectFZv () #13 0x000027ea in _Dmain () #14 0x000132df in D2rt6dmain24mainUiPPaZi7runMainMFZv () #15 0x00012e89 in D2rt6dmain24mainUiPPaZi7tryExecMFMDFZvZv () #16 0x00013327 in D2rt6dmain24mainUiPPaZi6runAllMFZv () #17 0x00012e89 in D2rt6dmain24mainUiPPaZi7tryExecMFMDFZvZv () #18 0x00012e23 in main () ---

The thread_get_state return code is 268435459 (0x10000003), probably corresponding to MACH_SEND_INVALID_DEST (mach/message.h). Just a shot in the dark, but maybe suspend() gets called for a thread that's no longer around?

And indeed, t.isRunning is false when thread_get_state() is called.

Fixed by Sean Kelly in https://github.com/D-Programming-Language/druntime/commit/45c4f1b94dddd091cc5aa72de7b0fddf76a6de21.

Reopened to be marked as fixed with the 2.054 release.

Closing, again, as 2.054 has been released in the meantime.