Segmentation fault on 0.6 & 0.7: deprecation warnings in threads

[awllee]

This may be quite low priority, since threads are experimental [although they work very well] and deprecations should be fixed. Nevertheless...

Deprecation warnings in threaded code appear to cause segmentation faults. This is much easier to reproduce if JULIA_NUM_THREADS=4, but the error can still occur with JULIA_NUM_THREADS=2. Replacing the deprecated method with its modern counterpart results in no segmentation faults.

using Compat.LinearAlgebra
import Compat.hasmethod
# no Compat.mul! or Compat.LinearAlgebra.mul!
if VERSION.minor == 6 mul! = A_mul_B! end

function foo1()
  v = method_exists(Base.sin, Tuple{Float64})
  # v = hasmethod(Base.sin, Tuple{Float64})
end

function foo2()
  A = [1.0 2.0; 3.0 4.0]
  b = [1.0;2.0]
  c = [0.0;0.0]
  A_mul_B!(c, A, b)
  # mul!(c, A, b)
end

function foo3()
  v = [1.0;2.0;3.0;4.0]
  log(v)
  # log.(v)
end

function bar(f::F) where F <: Function
  Threads.@threads for i in 1:4
    f()
  end
end

bar(foo1) # sometimes segfaults

bar(foo2) # often segfaults

# bar(foo3) # sometimes segfaults on v0.6.3

[yuyichao]

Dup of #14494