Consider using spawn and channel in specific async runtime.

[Berrysoft]

I see that EventStream is using std::thread and std::sync::mpsc at now. It is of course an improvement according to #448. However, I'd like to suggest a possible improvement, with the usage of spawn and channels in specific async runtime, for example, tokio or async-std. Therefore, the async runtime could manage the event stream thread.

The replacement in tokio is spawn and mpsc.

The replacement in async-std is spawn and channel.

Both of them should be drop-in replacement, rewritting some functions to async ones. I'd like to open a pull request later, trying to add the implementation as an optional feature, and keep the original implementation as a fallback one.

[BoolPurist]

I agree that the underlying stream should not used to spawn a naked OS-thread outside any async runtime.

The previous attempt to fix this seems to go back quite a while.

I am willing to pick this up from here.

However, I have a few questions before starting the rewrite.

1. Is still there now any interest in the rewrite ?
2. How to do it in terms of the compile features ? I have the following suggestion.

• Only supporting tokio under the flag "event-stream"
• The flag "event-stream" is working the context of tokio.
  One new compiler flag "asyn-std" is then using asyn-std.
• Two new flags "tokio" and "asyn-std" in which tokio and asyn-std are respectively used.
  In this variant, I would deprecate the flag "event-stream" and type EventStream simply
  since the implementation of this type with naked OS thread should not be used anymore
  going forward in my opinion.

[Berrysoft]

Is still there now any interest in the rewrite?

No, sorry. Now I think that it is easy enough for the users of tokio or other async runtimes to write another stream without spawing new threads.

[BoolPurist]

Is still there now any interest in the rewrite?

No, sorry. Now I think that it is easy enough for the users of tokio or other async runtimes to write another stream without spawing new threads.

Fair enough.
One question remains for me: shouldn't the usage of this feature then be discouraged/deprecated ?

[Berrysoft]

I think it may be better to add some comments in the doc about the spawnned thread, and suggest the user to write a runtime-specific one if they care about the overhead.

[BoolPurist]

I can go ahead and add this fact to the docs about the type EvenStream. While I am at it, I would also add those comments to the example with tokio or asyn-std using this type.