May 2011 - libdispatch-dev - lists.macosforge.org

Updates regarding the status of libdispatch on Windows
by Marius Zwicker 20 Jun '11

20 Jun '11

I am very glad to report good news on the topic libdispatch on windows. Yesterday we managed to eradicate the last obvious bugs and all tests and can now completely build and run libdispatch by using MSVC (Visual Studio 2010) on windows. All tests seem to pass or fail the same way they currently do on Linux, so at least we have reached feature parity ;) - Please see the test results attached [1], the responsible code can be found at http://opensource.mlba-team.de/svn/xdispatch/tags/0.4/ (see the core folder for the libdispatch implementation). The big question to raise now is wether merging back to macosforge.org is possible or not. We had to change a lot of the sources and some of the changes do not really contribute to the maintainability of the code, e.g. MSVC still does not support named intializers in structs. Additionally I split queue.c into several subfiles, such as debug.c and so on, to make the different implementations for the different platforms easier to find. As such it will be quite difficult to provide small patches instead of one or two bigger ones. And there is DrPizza/Peter's port as well, although I do not know how much of his work is tested. What do you think about it? - Marius Zwicker [1] Test results of xdispatch_tests.exe /k (which is a cross-platform implementation of the test-harness and tests in one executable): MUnit Test Framework with independent processes (c) 2011 MLBA Running all Tests ================================== Running test 'dispatch_api'... Passed (1) Running test 'dispatch_simpleFunction'... Passed (2) Running test 'dispatch_sync'... Passed (3) Running test 'dispatch_testDebug'... -> Printing object debugging information: -> Printing done. Passed (4) Running test 'dispatch_apply_function'... Passed (5) Running test 'dispatch_priority'... -> LOW: 0 -> DEFAULT: 64 **************************************** -> HIGH: 128 ******************************************************************************** Passed (6) Running test 'dispatch_priority2'... -> LOW: 64 **************************************** -> DEFAULT: 64 **************************************** -> HIGH: 64 **************************************** Passed (7) Running test 'dispatch_cascade'... -> maxcount = -2 * * * * * * * * * * * * *** * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * * * *** ** * ** * ** * ** * ** ** * * Passed (8) Running test 'dispatch_apply_serialqueue'... Passed (9) Running test 'dispatch_after'... -> must finish between 5.5s and 6.5s: 0.015614 -> must finish between 1.5s and 2.5s: 0.005205 -> must finish between 0.0s and 0.5s: 0.000000 Passed (10) Running test 'test_queue_finalizer'... Passed (11) Running test 'dispatch_starfish'... -> lap: 10 -> count: 300 -> delta: 27063 ns -> math: 0.149850 ns / lap -> lap: 9 -> count: 300 -> delta: 18223 ns -> math: 0.100903 ns / lap -> lap: 8 -> count: 300 -> delta: 19046 ns -> math: 0.105460 ns / lap -> lap: 7 -> count: 300 -> delta: 18235 ns -> math: 0.100969 ns / lap -> lap: 6 -> count: 300 -> delta: 18317 ns -> math: 0.101423 ns / lap -> lap: 5 -> count: 300 -> delta: 18339 ns -> math: 0.101545 ns / lap -> lap: 4 -> count: 300 -> delta: 18355 ns -> math: 0.101633 ns / lap -> lap: 3 -> count: 300 -> delta: 20610 ns -> math: 0.114120 ns / lap -> lap: 2 -> count: 300 -> delta: 21503 ns -> math: 0.119064 ns / lap -> lap: 1 -> count: 300 -> delta: 18919 ns -> math: 0.104756 ns / lap Passed (12) Running test 'dispatch_memory_use'... -> All workers done. Passed (13) Running test 'dispatch_group_function'... -> sleeping... -> sleeping... -> sleeping... -> done. -> done. -> done. Passed (14) Running test 'cross_blocks'... Passed (15) Running test 'dispatch_plusplus'... Passed (16) Running test 'dispatch_apply_blocks'... Passed (17) Running test 'dispatch_group_blocks'... -> sleeping... -> sleeping... -> sleeping... -> done. -> done. -> done. Passed (18) Running test 'dispatch_after_blocks'... -> must finish between 5.5s and 6.5s: 0.015614 -> must finish between 1.5s and 2.5s: 0.005205 -> must finish between 0s and .5s: 0.000000 Passed (19) Running test 'dispatch_pingpong_blocks'... Passed (20) Running test 'dispatch_semaphore'... Passed (21) Running test 'dispatch_timer_bit31'... Passed (22) Running test 'dispatch_timer_bit63'... -> 0 Passed (23) Running test 'dispatch_drift'... -> 1: jitter 0.000000, drift 0.000000 -> 2: jitter 0.006400, drift 0.006400 -> 3: jitter 0.012800, drift 0.006400 -> 4: jitter 0.003600, drift -0.009200 -> 5: jitter 0.010000, drift 0.006400 -> 6: jitter 0.000799, drift -0.009201 -> 7: jitter 0.007199, drift 0.006400 -> 8: jitter 0.013599, drift 0.006400 -> 9: jitter 0.004399, drift -0.009200 -> 10: jitter 0.010799, drift 0.006400 -> 11: jitter 0.001599, drift -0.009200 -> 12: jitter 0.007998, drift 0.006399 -> 13: jitter 0.014398, drift 0.006400 -> 14: jitter 0.005198, drift -0.009200 -> 15: jitter 0.011598, drift 0.006400 -> 16: jitter 0.002398, drift -0.009200 -> 17: jitter 0.008798, drift 0.006400 -> 18: jitter -0.000403, drift -0.009201 -> 19: jitter 0.005997, drift 0.006400 -> 20: jitter 0.012397, drift 0.006400 -> 21: jitter 0.003197, drift -0.009200 -> 22: jitter 0.009597, drift 0.006400 -> 23: jitter 0.000396, drift -0.009201 -> 24: jitter 0.006796, drift 0.006400 -> 25: jitter 0.013196, drift 0.006400 -> 26: jitter 0.003996, drift -0.009200 -> 27: jitter 0.010396, drift 0.006400 -> 28: jitter 0.001196, drift -0.009200 -> 29: jitter 0.007595, drift 0.006399 -> 30: jitter 0.013995, drift 0.006400 -> 31: jitter 0.004795, drift -0.009200 -> 32: jitter 0.011195, drift 0.006400 -> 33: jitter 0.001995, drift -0.009200 -> 34: jitter 0.008395, drift 0.006400 -> 35: jitter 0.014794, drift 0.006399 -> 36: jitter 0.005594, drift -0.009200 -> 37: jitter 0.011994, drift 0.006400 -> 38: jitter 0.002794, drift -0.009200 -> 39: jitter 0.009194, drift 0.006400 -> 40: jitter -0.000007, drift -0.009201 -> 41: jitter 0.006393, drift 0.006400 -> 42: jitter 0.012793, drift 0.006400 -> 43: jitter 0.003593, drift -0.009200 -> 44: jitter 0.009993, drift 0.006400 -> 45: jitter 0.000793, drift -0.009200 -> 46: jitter 0.007192, drift 0.006399 -> 47: jitter 0.013592, drift 0.006400 -> 48: jitter 0.004392, drift -0.009200 -> 49: jitter 0.010792, drift 0.006400 -> 50: jitter 0.001592, drift -0.009200 -> 51: jitter 0.007992, drift 0.006400 -> 52: jitter 0.014391, drift 0.006399 -> 53: jitter 0.005191, drift -0.009200 -> 54: jitter 0.011591, drift 0.006400 -> 55: jitter 0.002391, drift -0.009200 -> 56: jitter 0.008791, drift 0.006400 -> 57: jitter -0.000410, drift -0.009201 -> 58: jitter 0.005990, drift 0.006400 -> 59: jitter 0.012390, drift 0.006400 -> 60: jitter 0.003190, drift -0.009200 -> 61: jitter 0.009590, drift 0.006400 -> 62: jitter 0.000390, drift -0.009200 -> 63: jitter 0.006789, drift 0.006399 -> 64: jitter 0.013189, drift 0.006400 -> 65: jitter 0.003989, drift -0.009200 -> 66: jitter 0.010389, drift 0.006400 -> 67: jitter 0.001189, drift -0.009200 -> 68: jitter 0.007589, drift 0.006400 -> 69: jitter 0.013988, drift 0.006399 -> 70: jitter 0.004788, drift -0.009200 -> 71: jitter 0.011188, drift 0.006400 -> 72: jitter 0.001988, drift -0.009200 -> 73: jitter 0.008388, drift 0.006400 -> 74: jitter 0.014788, drift 0.006400 -> 75: jitter 0.005587, drift -0.009201 -> 76: jitter 0.011987, drift 0.006400 -> 77: jitter 0.002787, drift -0.009200 -> 78: jitter 0.009187, drift 0.006400 -> 79: jitter -0.000013, drift -0.009200 -> 80: jitter 0.006386, drift 0.006399 -> 81: jitter 0.012786, drift 0.006400 -> 82: jitter 0.003586, drift -0.009200 -> 83: jitter 0.009986, drift 0.006400 -> 84: jitter 0.000786, drift -0.009200 -> 85: jitter 0.007186, drift 0.006400 -> 86: jitter 0.013585, drift 0.006399 -> 87: jitter 0.004385, drift -0.009200 -> 88: jitter 0.010785, drift 0.006400 -> 89: jitter 0.001585, drift -0.009200 -> 90: jitter 0.007985, drift 0.006400 -> 91: jitter 0.014385, drift 0.006400 -> 92: jitter 0.005184, drift -0.009201 -> 93: jitter 0.011584, drift 0.006400 -> 94: jitter 0.002384, drift -0.009200 -> 95: jitter 0.008784, drift 0.006400 -> 96: jitter -0.000416, drift -0.009200 -> 97: jitter 0.005983, drift 0.006399 -> 98: jitter 0.012383, drift 0.006400 -> 99: jitter 0.003183, drift -0.009200 -> 100: jitter 0.009583, drift 0.006400 -> 101: jitter 0.000383, drift -0.009200 -> 102: jitter 0.006783, drift 0.006400 -> 103: jitter 0.013182, drift 0.006399 -> 104: jitter 0.003982, drift -0.009200 -> 105: jitter 0.010382, drift 0.006400 -> 106: jitter 0.001182, drift -0.009200 -> 107: jitter 0.007582, drift 0.006400 -> 108: jitter 0.013982, drift 0.006400 -> 109: jitter 0.004781, drift -0.009201 -> 110: jitter 0.011181, drift 0.006400 -> 111: jitter 0.001981, drift -0.009200 -> 112: jitter 0.008381, drift 0.006400 -> 113: jitter 0.014781, drift 0.006400 -> 114: jitter 0.005580, drift -0.009201 -> 115: jitter 0.011980, drift 0.006400 -> 116: jitter 0.002780, drift -0.009200 -> 117: jitter 0.009180, drift 0.006400 -> 118: jitter -0.000020, drift -0.009200 -> 119: jitter 0.006380, drift 0.006400 -> 120: jitter 0.012779, drift 0.006399 -> 121: jitter 0.003579, drift -0.009200 -> 122: jitter 0.009979, drift 0.006400 -> 123: jitter 0.000779, drift -0.009200 -> 124: jitter 0.007179, drift 0.006400 -> 125: jitter 0.013579, drift 0.006400 -> 126: jitter 0.004378, drift -0.009201 -> 127: jitter 0.010778, drift 0.006400 -> 128: jitter 0.001578, drift -0.009200 -> 129: jitter 0.007978, drift 0.006400 -> 130: jitter 0.014378, drift 0.006400 -> 131: jitter 0.005178, drift -0.009200 -> 132: jitter 0.011577, drift 0.006399 -> 133: jitter 0.002377, drift -0.009200 -> 134: jitter 0.008777, drift 0.006400 -> 135: jitter -0.000423, drift -0.009200 -> 136: jitter 0.005977, drift 0.006400 -> 137: jitter 0.012376, drift 0.006399 -> 138: jitter 0.003176, drift -0.009200 -> 139: jitter 0.009576, drift 0.006400 -> 140: jitter 0.000376, drift -0.009200 -> 141: jitter 0.006776, drift 0.006400 -> 142: jitter 0.013176, drift 0.006400 -> 143: jitter 0.003975, drift -0.009201 -> 144: jitter 0.010375, drift 0.006400 -> 145: jitter 0.001175, drift -0.009200 -> 146: jitter 0.007575, drift 0.006400 -> 147: jitter 0.013975, drift 0.006400 -> 148: jitter 0.004775, drift -0.009200 -> 149: jitter 0.011174, drift 0.006399 -> 150: jitter 0.001974, drift -0.009200 -> 151: jitter 0.008374, drift 0.006400 -> 152: jitter 0.014774, drift 0.006400 -> 153: jitter 0.005574, drift -0.009200 -> 154: jitter 0.011973, drift 0.006399 -> 155: jitter 0.002773, drift -0.009200 -> 156: jitter 0.009173, drift 0.006400 -> 157: jitter -0.000027, drift -0.009200 -> 158: jitter 0.006373, drift 0.006400 -> 159: jitter 0.012773, drift 0.006400 -> 160: jitter 0.003572, drift -0.009201 -> 161: jitter 0.009972, drift 0.006400 -> 162: jitter 0.000772, drift -0.009200 -> 163: jitter 0.007172, drift 0.006400 -> 164: jitter 0.013572, drift 0.006400 -> 165: jitter 0.004372, drift -0.009200 -> 166: jitter 0.010771, drift 0.006399 -> 167: jitter 0.001571, drift -0.009200 -> 168: jitter 0.007971, drift 0.006400 -> 169: jitter 0.014371, drift 0.006400 -> 170: jitter 0.005171, drift -0.009200 -> 171: jitter 0.011570, drift 0.006399 -> 172: jitter 0.002370, drift -0.009200 -> 173: jitter 0.008770, drift 0.006400 -> 174: jitter -0.000430, drift -0.009200 -> 175: jitter 0.005970, drift 0.006400 -> 176: jitter 0.012370, drift 0.006400 -> 177: jitter 0.003169, drift -0.009201 -> 178: jitter 0.009569, drift 0.006400 -> 179: jitter 0.000369, drift -0.009200 -> 180: jitter 0.006769, drift 0.006400 -> 181: jitter 0.013169, drift 0.006400 -> 182: jitter 0.003969, drift -0.009200 -> 183: jitter 0.010368, drift 0.006399 -> 184: jitter 0.001168, drift -0.009200 -> 185: jitter 0.007568, drift 0.006400 -> 186: jitter 0.013968, drift 0.006400 -> 187: jitter 0.004768, drift -0.009200 -> 188: jitter 0.011168, drift 0.006400 -> 189: jitter 0.001967, drift -0.009201 -> 190: jitter 0.008367, drift 0.006400 -> 191: jitter 0.014767, drift 0.006400 -> 192: jitter 0.005567, drift -0.009200 -> 193: jitter 0.011967, drift 0.006400 -> 194: jitter 0.002766, drift -0.009201 -> 195: jitter 0.009166, drift 0.006400 -> 196: jitter -0.000034, drift -0.009200 -> 197: jitter 0.006366, drift 0.006400 -> 198: jitter 0.012766, drift 0.006400 -> 199: jitter 0.003566, drift -0.009200 -> 200: jitter 0.009965, drift 0.006399 -> 201: jitter 0.000765, drift -0.009200 -> 202: jitter 0.007165, drift 0.006400 -> 203: jitter 0.013565, drift 0.006400 -> 204: jitter 0.004365, drift -0.009200 -> 205: jitter 0.010765, drift 0.006400 -> 206: jitter 0.001564, drift -0.009201 -> 207: jitter 0.007964, drift 0.006400 -> 208: jitter 0.014364, drift 0.006400 -> 209: jitter 0.005164, drift -0.009200 -> 210: jitter 0.011564, drift 0.006400 -> 211: jitter 0.002363, drift -0.009201 -> 212: jitter 0.008763, drift 0.006400 -> 213: jitter -0.000437, drift -0.009200 -> 214: jitter 0.005963, drift 0.006400 -> 215: jitter 0.012363, drift 0.006400 -> 216: jitter 0.003163, drift -0.009200 -> 217: jitter 0.009562, drift 0.006399 -> 218: jitter 0.000362, drift -0.009200 -> 219: jitter 0.006762, drift 0.006400 -> 220: jitter 0.013162, drift 0.006400 -> 221: jitter 0.003962, drift -0.009200 -> 222: jitter 0.010362, drift 0.006400 -> 223: jitter 0.001161, drift -0.009201 -> 224: jitter 0.007561, drift 0.006400 -> 225: jitter 0.013961, drift 0.006400 -> 226: jitter 0.004761, drift -0.009200 -> 227: jitter 0.011161, drift 0.006400 -> 228: jitter 0.001960, drift -0.009201 -> 229: jitter 0.008360, drift 0.006400 -> 230: jitter 0.014760, drift 0.006400 -> 231: jitter 0.005560, drift -0.009200 -> 232: jitter 0.011960, drift 0.006400 -> 233: jitter 0.002760, drift -0.009200 -> 234: jitter 0.009159, drift 0.006399 -> 235: jitter -0.000041, drift -0.009200 -> 236: jitter 0.006359, drift 0.006400 -> 237: jitter 0.012759, drift 0.006400 -> 238: jitter 0.003559, drift -0.009200 -> 239: jitter 0.009959, drift 0.006400 -> 240: jitter 0.000758, drift -0.009201 -> 241: jitter 0.007158, drift 0.006400 -> 242: jitter 0.013558, drift 0.006400 -> 243: jitter 0.004358, drift -0.009200 -> 244: jitter 0.010758, drift 0.006400 -> 245: jitter 0.001557, drift -0.009201 -> 246: jitter 0.007957, drift 0.006400 -> 247: jitter 0.014357, drift 0.006400 -> 248: jitter 0.005157, drift -0.009200 -> 249: jitter 0.011557, drift 0.006400 -> 250: jitter 0.002357, drift -0.009200 Failed Test average jitter: fabs(*jittersum) / (double)(*count) < 0.0001 failed (was 0.007169,0.000100), at: ..\..\..\tests\dispatch_drift.cpp:94 (24) Running test 'cxx_dispatch_debug'... -> Begin testing debug output using std::cout xdispatch::queue (com.apple.main-thread) xdispatch::group xdispatch::semaphore -> All output done. (25) Running test 'cxx_dispatch_queue'... Passed (26) Running test 'cxx_dispatch_mainqueue'... Passed (27) Running test 'cxx_dispatch_cascade'... Passed (28) Running test 'cxx_dispatch_group'... -> sleeping... -> sleeping... -> sleeping... -> done. -> done. -> done. Passed (29) Running test 'cxx_dispatch_serialqueue'... Passed (30) Running test 'cxx_dispatch_semaphore'... Passed (31) Running test 'cxx_dispatch_current'... -> Queue should be global default queue xdispatch::queue (com.apple.root.default-priority) -> Queue should be main queue xdispatch::queue (com.apple.main-thread) Passed (32) Running test 'cxx_dispatch_timer'... -> Testing single-shot timer -> Testing periodic timer -> 0 -> 1 -> 2 -> 3 -> 4 -> 5 Passed (33) Running test 'Qt_runnable'... Passed (34) Running test 'Qt_dispatch_debug'... -> Begin testing debug output using QDebug() -> All output done. (35) Running test 'Qt_dispatch_queue'... Passed (36) Running test 'Qt_dispatch_mainqueue'... -> System usage should be zero Passed (37) Running test 'Qt_dispatch_cascade'... Passed (38) Running test 'Qt_dispatch_after'... -> Should finish between 1s and 1.5s: 0.998000 Failed Test watch.elapsed()>=1000 failed (was 0,0), at: ..\..\..\tests\Qt_dispatch_timeout.cpp:48 (39) Running test 'Qt_dispatch_group'... -> signal 'all_finished' emitted Passed (40) Running test 'Qt_dispatch_serialqueue'... Passed (41) Running test 'Qt_dispatch_semaphore'... Passed (42) Running test 'Qt_dispatch_current'... -> Queue should be global default queue -> Queue should be main queue Passed (43) Running test 'Qt_synchronized'... -> Testing 'synchronized' keyword (apply) -> 0.000160 ms per Iteration -> Testing 'synchronized' keyword (for / group) -> 0.000000 ms per Iteration -> Testing 'synchronize' keyword (apply) -> 0.000160 ms per Iteration -> Testing 'synchronize' keyword (for / group) -> 0.000000 ms per Iteration -> Testing with normal mutex (apply) -> 0.002180 ms per Iteration Passed (44) ================================== !!! NOT ALL PASSED !!! ---------------------------------- Passed: 42 Failed: 2

7 19

[PATCH] Quiet the warning of unused return value
by Joakim Johansson 28 May '11

28 May '11

I get a warning as the return value is not checked from dispatch_atomic_cmpxchg() - as it is in all other 11 usages of it. Attached is a patch that just discards the return value accordingly to quiet this warning. Joakim Index: src/queue.c =================================================================== --- src/queue.c (revision 197) +++ src/queue.c (working copy) @@ -377,7 +377,7 @@ // The first xchg on the tail will tell the enqueueing thread that it // is safe to blindly write out to the head pointer. A cmpxchg honors // the algorithm. - dispatch_atomic_cmpxchg(&dq->dq_items_head, mediator, NULL); + (void) dispatch_atomic_cmpxchg(&dq->dq_items_head, mediator, NULL); _dispatch_debug("no work on global work queue"); return NULL; }

2 1

[PATCH] bug fix for invalid comparison with unsigned value < 0
by Joakim Johansson 28 May '11

28 May '11

The comparison "_dispatch_hw_config.cc_max_active < 0” is of an unsigned value and thus never true, the attached patch uses and checks failure of sysconf() properly instead. Cheers, Joakim Index: src/queue.c =================================================================== --- src/queue.c (revision 197) +++ src/queue.c (working copy) @@ -474,12 +474,13 @@ _dispatch_hw_config.cc_max_physical = _dispatch_hw_config.cc_max_active; #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN) - _dispatch_hw_config.cc_max_active = (int)sysconf(_SC_NPROCESSORS_ONLN); - if (_dispatch_hw_config.cc_max_active < 0) - _dispatch_hw_config.cc_max_active = 1; + int ret; + + ret = (int)sysconf(_SC_NPROCESSORS_ONLN); + _dispatch_hw_config.cc_max_logical = _dispatch_hw_config.cc_max_physical = - _dispatch_hw_config.cc_max_active; + _dispatch_hw_config.cc_max_active = (ret < 0) ? 1 : ret; #else #warning "_dispatch_queue_set_width_init: no supported way to query CPU count" _dispatch_hw_config.cc_max_logical =

1 0

libdispatch object caches
by Joakim Johansson 25 May '11

25 May '11

Hi, I have a question about a possible optimization and would appreciate feedback on whether it would be considered a) worthwhile to investigate b) a candidate for integration if providing good results. The reason for asking is that I’ve seen significantly better (seems to be around 2x) performance on Solaris for e.g. ‘dispatch starfish’ for the lap times, which seems to be malloc-bound (as is also commented in the starfish implementation). libdispatch today uses optimized TSD dispatch continuation object caches with a heavily optimized implementation for Darwin. This cache seems unbounded during the lifetime of a given drain of the queue, after which a forced cleanup is performed (by calling _dispatch_force_cache_cleanup()). The proposed change would be to let _dispatch_continuation_alloc_from_heap() allocate its objects from a magazine based object cache, and let _dispatch_cache_cleanup2() to return objects back to that object cache. This is the approach taken in libumem (see http://www.usenix.org/event/usenix01/full_papers/bonwick/bonwick.pdf for background) - in fact, the easiest approach would possibly be to even depend on libumem being available (https://labs.omniti.com/trac/portableumem exists for non-Solaris platforms) - this could be compile-time optional of course depending on the availability of libumem in such a case, but that would implicitly replace the default memory allocator (if we don’t ‘cherry pick’ out e.g. the magazine layer). It seems that this would allow us to reuse dispatch continuation objects across invocations/threads in an efficient way. So there are a few questions: - Would a magazine-based object cache be interesting for libdispatch? - Have any analysis of typical TSD continuation cache sizes been done already? (anyone who have seen any issues here in GCD-heavy applications?) - What would be a reasonable test for performance validation? - Would a relying on libumem be acceptable for people in general? One other possibility would be to only keep a single item in the TSD and fall back on such an object-cache if that single object would not be available (that is, releasing it back to the object pool in _dispatch_continuation_free, the ‘hot continuation’ trick from _dispatch_continuation_pop() would thus still work) - but this would require more serious investigation (the magazine approach above would be a ’safer’ change). Thoughts? Joakim PS Apple folks also have rdar://4944235 for libumem/OS X, as it would be a nice analogue to the GCD system-level thread management, but for the memory allocation / object cacheing subsystems instead

3 2

[patch] workaround for GNU/Linux pthread_exit() bug
by Mark Heily 17 May '11

17 May '11

See below for a patch to dispatch_main() that is needed on GNU/Linux. This platform cannot call pthread_exit() from the main thread because of an undocumented bug that causes the process to become a zombie. As an alternative, this patch manually calls the TSD destructors, blocks all signals, and parks the main thread using pause(). If there is anything else that needs to be done to make this a viable alternative to calling pthread_exit(), please let me know. Regards, - Mark $ svn di src/queue.c Index: src/queue.c =================================================================== --- src/queue.c (revision 197) +++ src/queue.c (working copy) @@ -941,8 +941,32 @@ if (pthread_main_np()) { #endif _dispatch_program_is_probably_callback_driven = true; +#if defined(__linux__) + /* + * Workaround for a GNU/Linux bug that causes the process to + * become a zombie when the main thread calls pthread_exit(). + */ + sigset_t mask; + void *p; + + p = pthread_getspecific(dispatch_queue_key); + if (p != NULL) + _dispatch_queue_cleanup(p); + p = pthread_getspecific(dispatch_sema4_key); + if (p != NULL) + _dispatch_release(p); + p = pthread_getspecific(dispatch_cache_key); + if (p != NULL) + _dispatch_cache_cleanup2(p); + + sigfillset(&mask); + pthread_sigmask(SIG_SETMASK, &mask, NULL); + pause(); + DISPATCH_CRASH("pause() returned"); +#else /* !defined(__linux__) */ pthread_exit(NULL); DISPATCH_CRASH("pthread_exit() returned"); +#endif #if HAVE_PTHREAD_MAIN_NP } DISPATCH_CLIENT_CRASH("dispatch_main() must be called on the main thread");

2 1