On my fairly powerful machine, I can reliably segfault the Fibonacci test when changing n to 16.
n = 15 sometimes work and sometimes segfault.
The issue seems to be in libconcurrent, the backtrace is quite long so I'm pasting just the end:
heap 11 is full
thread 2 resuming task 0x7ffff7b9f590
thread 2 task 0x7ffff7b9f590 sync on task 0x7ffff57a5290
thread 2 resuming task 0x7ffff7b85990
thread 2 completed task 0x7ffff7b85990
thread 2 adding from task 0x7ffff7b85990's CQ: 0x7ffff7ba0e10
thread 2 resuming task 0x7ffff7ba4710
thread 2 completed task 0x7ffff7ba4710
thread 2 adding from task 0x7ffff7ba4710's CQ: 0x7ffff7ba4190
thread 2 resuming task 0x7ffff57a1d90
thread 2 task 0x7ffff57a1d90 spawned task 0x7ffff7b85c10
thread 2 had task 0x7ffff57a1d90 yield
heap 11 is full
thread 2 resuming task 0x7ffff7b82390
thread 2 task 0x7ffff7b82390 sync on task 0x7ffff57a1890
thread 2 resuming task 0x7ffff7b81b10
thread 2 task 0x7ffff7b81b10 sync on task 0x7ffff57a2190
thread 2 resuming task 0x7ffff7ba0e10
thread 2 task 0x7ffff7ba0e10 sync on task 0x7ffff7ba2190
Thread 1 "fib" received signal SIGSEGV, Segmentation fault.
0x000055555555a37a in partr_sync ()
(gdb) bt
#0 0x000055555555a37a in partr_sync ()
#1 0x000055555555676d in fib (start=<optimized out>, end=<optimized out>, arg_=0x2) at fib.c:14
#2 fib (arg_=0x2, start=<optimized out>, end=<optimized out>) at fib.c:5
#3 0x000055555555827b in partr_coro ()
#4 0x000055555555c71b in concurrent_arch_trampoline_to_caller ()
#5 0x0000555555591e40 in ?? ()
#6 0x0000555555591e40 in ?? ()
#7 0x0000000000000000 in ?? ()
Note that Julia 1.3.0-rc4 does not exhibit this issue (but has scalability issues, will open another issue on Julia repo).