Fix for OSX spawn test

[quinnj]

Ran into this several times while running the tests about a million times so far for #22441. I don't think this is related to #17626, but one could always hope, right?

[iblislin]

Not sure it's related or not, I found a way to make success(`true`) insane.

1. borrow a MBA from my classmate :p
2. test.jl

using Base.Test

for i in 1:100000  # just a large number
  println(i)
  @test success(`true`)
end

yes, we need @test to trigger crashing/freezing. no idea why single success works fine.
3. julia test.jl
4. hold your Crtl+t, keep sending SIGINFO
5. sometimes you got segmentation fault; sometimes the program freezed, only kill -9 can stop it.

[vtjnash]

@iblis17 looks unrelated. I ran your code in my debugger and found the issue is that Apple's libunwind does not conform to its specification (its unw_init_local allocates memory, but is documented to not allocate memory)

[iblislin]

@vtjnash :-o file it as another issue?

[vtjnash]

Should be a radar, but I'm not very active about getting those submitted.