pike.git/ src/ testsuite.in

2011-04-02

2011-04-02 12:55:21 by Martin Stjernholm <mast@lysator.liu.se>

• 52500d4fe5e7da6a7ce09c23446e15a28550db1c (10 lines) (+10/-0) ^{[ Show | Annotate ]}
  Branch: 7.9

Disabled the bothersome yield test.

Afaics the test fails for the intended reason, but the actual yielding
doesn't work any better than this. The use of th_yield needs to be
replaced with some more controlled thread switching method.

5056:   dnl this will crash pike on out-of-address-space-related errors   dnl test_any([[ catch { allocate(10000,thread_create)(lambda() { sleep(1); })->wait(); } ]])    + cond(0, [[    test_any([[    // Test that a loop relinquishes the interpreter lock every now and then. -  +  // +  // This test is unfortunately disabled since it fails too often. +  // The subthread does really yield in the loop, but the main +  // thread, which really is waiting, fails to run. My conclusion is +  // that the problem is the unreliability of th_yield() - we need +  // to replace it with something else. :( +  // /mast    int count, stop, fail;    Thread.thread_create (lambda() {    int t_start = time();
5076:    sleep (0.1);    return fail ? -1 : count > 0;    ]], 1) + ]])       // _disable_threads    test_any([[