Added windows support for SharedArray

[twadleigh]

I made every effort to implement this with the least amount of change to existing code. This includes trying to maximize reuse of the existing windows version of the mmap_array function. One not-so-pleasing side effect of this is a clunkier looking signature with a Union type and a couple of new if-else blocks in the body.

I verified that test/parallel.jl passes on linux-64, windows-32, and windows-64 on my machine.

[amitmurthy]

Awesome. LGTM. Will merge in a day if there are no concerns raised.

cc: @vtjnash

[vtjnash]

missing utf16 call? or perhaps this is handled automatically now?

[tkelman]

line 141

[vtjnash]

ah. good call

[ViralBShah]

@amitmurthy Just bumping this - You mentioned someone who needed this capability on windows.

[amitmurthy]

This should backport cleanly, however, we should let it bake in this branch for a while before doing so.

Adding a @JuliaBackports just for reference.

[timholy]

Yay, this is great! Thanks, @twadleigh!

[twadleigh]

@timholy, I'm very gratified to have been able to contribute something useful to base.

[tkelman]

would prefer testing it for a while on master before backporting, 0.3.3 is scheduled for pretty soon

[amitmurthy]

That's what I meant.

What is the protocol for adding a @juliabackports and backport_pending label? I added them just to track this commit.

[tkelman]

both the notification and the label are useful, but for commits that aren't attached to a PR we can only use the notification.

[tkelman]

The backport window is open now for 0.3.4-pre. This doesn't cherry-pick cleanly and should be done carefully due to things like AbstractString and UInt capitalization. This should be okay to backport since it seems to be working pretty well on master, but maybe would be best if @twadleigh could prepare a PR against release-0.3 for the backported version due to the conflicts.

[tkelman]

thanks @twadleigh for preparing the backport. removing the label from here since it's now covered by #9148