promotion to SharedArray

[skariel]

so the following code eventually produces a non-shared array and for me it was a surprise:

a = SharedArray(Float32, 3)
a[:] = 0
a += 1

I have a function that sometimes gets an Array and other times a SharedArray. Now it has a different behaviour depending on the type used.

Is this behaviour intended?

[KristofferC]

Ref #12964

[skariel]

(a quick fix of course is to iterate and do everything in-place to keep it general. Using a.s += 1 will make it not general)

[JohnHBrock]

@skariel Is this on master? Can you run versioninfo() in the REPL and paste what it outputs?

[skariel]

Julia Version 0.4.3
Commit a2f713d (2016-01-12 21:37 UTC)
Platform Info:
  System: Linux (x86_64-unknown-linux-gnu)
  CPU: Intel(R) Core(TM) i7-4700HQ CPU @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

[DanielArndt]

This is very related to the segfault I reported in #14459 -- But as you can see by the reference @KristofferC gave, it appears this was an intentional change.

[tkelman]

Not an uncontroversial one though.