Fix Windows onExit failure.

[jamestalmage]

SIGHUP seems to be the only kill signal supported by Windows:
https://nodejs.org/api/process.html#process_process_kill_pid_signal

This is a stab in the dark while I wade through setting up Node on Windows.

[jamestalmage]

😢 If only it were that easy.

[sindresorhus]

You're dealing with computers. It's never easy :p

[sindresorhus]

You can just use childProcess.execFile(process.execPath, ['../cli.js'].concat(args), {cwd: __dirname}, cb); instead of which.

[jamestalmage]

Finally!
@sindresorhus - I'm gonna let the latest push go green and then squash

[jamestalmage]

I might have actually fixed this line too.

Do we have a test that covers this?

Everything passed on OSX / Node 4.1 without this and without my change (in other words, unless the problem this addresses is specific to an OS / Node version we do NOT have test coverage for it).

[jamestalmage]

@vdemedes, this line is from #47

Can you remember what you were addressing there? Tests pass without it. (even before my changes).

[vadimdemedes]

process.stdout.write(''); was there before me, @sindresorhus should have a reason for it.

[jamestalmage]

oops. Sorry - blame attributed it to you, and I didn't scan down the #47 far enough.

[jamestalmage]

@sindresorhus ?

[sindresorhus]

Yeah, I added it as a quick hack because the 1 test passed message wouldn't output without it when running multiple tests and I couldn't reproduce it in a test. If it works fine now you can just remove it, things have changed so much since then.

[jamestalmage]

This means the test process is now receiving messages.

Do we need to monkey patch process.on so tests never see them?

[sindresorhus]

Not unless it interferes with them. @vdemedes ?

[jamestalmage]

It won't interfere with AVA test code.

But if someone attaches a process.on('message', handler), and then emits their own message events on process, that would interfere. Though it would be really odd behavior.

[jamestalmage]

It wouldn't really ever make sense for them to do that though (not unless they fork an additional process).

[jamestalmage]

@sindresorhus @vdemedes
I've made the recommended changes. Anything else holding this up? It would be nice if peoples PR's could go green again.

[sindresorhus]

Landed! Superb work tracking this down @jamestalmage. Node.js 0.10 is getting so annoying to support...

[vadimdemedes]

I just realized something. Why don't we just let forks exit themselves, without interaction with a parent process?

process.send(results);
process.exit();

[jamestalmage]

I don't think there is anything stopping us from doing that now.

I did have one thought with regard to #140, and that is to leave processes running as long as possible to give us the best chance of catching extra assertion calls (beyond plan), and repetitive calls to t.end()

In that case it would be up to the parent to close.

But if we think that is a bad idea, then no - there is probably no reason to do it the way we are.