Add windows support

[avdi]

I'm starting to do more screencasts on the Windows side, so this is kind of a big deal for me.

[JoshCheek]

Oh, since you spent some time in there, I'm not sure if you found a good way to hit a given problem test, but one that's worked well for me is to add a tag to the spec, and running just that tag. eg: it("...", t:true) { ... } then I can run it with rspec -t t

I like it better than specifying the lineno b/c I don't have to pluck the line number out of a sea of text, and it keeps working even if I edit the test file in a way that moves the lines around.

It is a bit cavalier b/c it ignores bundler, but it keeps the pace quick (if you're more patient than me, then this might be a poor tradeoff). If you do hit issues from ignoring bundler, though, something like $ ruby $(rake require_paths) -S rspec -t t should work (bundler is just too expensive when working with subprocesses this extensively).

[avdi]

Commentary on all of the above as soon as I have time :-)

[JoshCheek]

• Looks it leaves orphans if we don't call stop, so child.stop needs to go in the ensure block as well. It didn't fail the test b/c it was the one that stubbed Process.kill to receive "-INT" looks like I tried to get some "for real" tests on that behaviour, but they all used fork >.<
• It looks like you can pass a port of 0, and it will be dynamically allocated (ie no risk of two SiB processes interfering with each other by binding to the same port). You can get the dynamic port with server.addr[1], would you mind trying this on Windows for me?
• Is there value in exposing port at the external interface? Given the volatility of the parent/child protocol, I'm hesitant to expose any of that info at a higher level. Eg if I'd allowed the user to pass the file descriptor, then it'd be difficult for this PR to switch to TCP.

[avdi]

I'm going to go point-by-point on these as soon as I can, but I wanted to start things off with a backgrounder: https://www.rubytapas.com/2016/12/14/ruby-code-on-windows/

[avdi]

I like the idea of a dynamic port a lot. I was wondering about that myself last night after I knocked off for the night.

[JoshCheek]

Hey, @avdi, I'm having difficulty getting the ChildProcess gem to work in both environments. If we don't call child.stop in the ensure block, it leaves orphan processes. But when we do call child.stop, windows gives me an error "The handle is invalid".

I've recreated it in isolation here and it's passing on Travis but failing on AppVeyor. Once those tests pass on both, I should be able to translate it for SiB.

Any chance you could try getting it working on Windows since you have native access to that environment? (I got the VM downloaded and running in VirtualBox, but my computer doesn't seem to be able to handle it, it quickly freezes up on me). If so, I can add you as a collaborator.

[JoshCheek]

I opened an issue on ChildProcess, too: https://github.com/enkessler/childprocess/issues/108

[JoshCheek]

Tried again and I got it cloned and ran the tests. Looks like the VM will work for me if I stay off the internet. I'll try looking at it tonight.

[JoshCheek]

Okay, got all the tests in the experiment passing on Windows and Unix. It's tricky, but I think I know the trick now: on Windows, you can't call stop after calling wait unless it interrupts the wait (or the "handle" will be invalid -- note that I don't know what a handle is), on Unix, you must call stop no matter what (or it will leave orphans).

[KAGsundaram]

For what it's worth, the blog post which @avdi references several times now has a few more reference links. https://www.rubytapas.com/2016/12/14/ruby-code-on-windows/