Don't wait for a chan that will never close

[MarcoPolo]

Fixes a subtle bug where we assumed we would always have a background task when we have a trace object. This would cause trace.Close() to hang indefinitely.

The fix is to use a waitgroup instead of a semaphore channel. Then we add to the waitgroup when we spawn a background task.

[gammazero]

LGTM.

You could also create the channel at the point the where you add to the WaitGroup, and then in Close check if the channel is not nil before waiting on it.

[MarcoPolo]

LGTM.

You could also create the channel at the point the where you add to the WaitGroup, and then in Close check if the channel is not nil before waiting on it.

Good to know! Are there any advantages compared to using a waitgroup?

My reasoning with the waitgroup is that if another future trace/reporter wants to do a background task they can do so easily by incrementing a waitgroup rather than managing multiple channels (or doing this refactor).

[gammazero]

Are there any advantages compared to using a waitgroup?

It is more about conveying meaning, with the channel it clearly says you are waiting on one thing. I don't think it really matters here.