worker: fix simpleworker Kill

worker: fix simpleworker Kill

gccgo does not like it when you
close a channel twice. It's not
a very nice thing anyway.

Fixes lp:1308852

https://code.launchpad.net/~axwalk/juju-core/lp1308852-worker-simple-gccgo/+merge/220563

(do not edit description out of merge proposal)

[axw, 2014-05-22 05:29:49]

Please take a look.

[wallyworld, 2014-05-22 06:13:30]

LGTM

[rog, 2014-05-22 08:38:34]

https://codereview.appspot.com/98480046/diff/1/worker/simpleworker.go
File worker/simpleworker.go (right):

https://codereview.appspot.com/98480046/diff/1/worker/simpleworker.go#newcode35
worker/simpleworker.go:35: case w.stopc <- struct{}{}:
This isn't quite right because it changes the NewSimpleWorker contract, which
states that stopCh is closed when the worker is killed, which means that we can
read from the channel any number of times after it's been closed.

If we want to avoid the recover check (which is something that gccgo really
should allow, but that's a different issue) the correct solution is to have a
mutex that guards the close and do something like this:

w.killMutex.Lock()
defer w.killMutex.Unlock()
select {
case <-w.stopc:
    return  // stopc is already closed
default:
    close(w.stopc)
}