Possible race condition in core.thread (at least on FreeBSD/x86_64 and Linux/PPC64)

[redstar]

On FreeBSD/x86_64 and Linux/PPC64, the test "Multiple threads running shared fibers" of module core.thread segfaults.

The observed behaviour is:

• the threads are created but they don't start
• the main thread calls joinAll
• the first thread starts
• a fiber context switch is initiated

=> boom!

The segfault vanishes if you change the timings, e.g. insert a `printf´ into the thread delegate.

This does not happen on Linux/x86 and Linux/x86_64. Therefore it is unclear if it is a race condition in the unit test or a code generation bug.

The crash on OS X reported in issue #492 might have the same root cause.

[redstar]

Same failure on mingw32 with release 0.13.0. Requires compiling with at least -O2.