Make it possible to use multiple websockets and callbacks

[edennis]

As described in #33 the internal state of the effects manager wasn't equipped to handle callbacks for multiple websockets. Using a tuple of the Endpoint and Ref now makes that possible.

I also had to fix the broken documentation in Phoenix.Presence to make elm-make stop complaining. The extra whitespace in the diff is due to elm-format.

[markhamburg]

This will be more efficient in general if the callback dictionary is keyed as ( ref, endpoint ) instead of ( endpoint, ref ) because it is more likely that the refs will be different than it is that the endpoints will be different. But otherwise, this looks correct.

[edennis]

I changed the ordering in the tuple to optimize comparisons in the Dict. For the other readers, the relevant code that handles comparisons in Elm is: https://github.com/elm-lang/core/blob/5.1.1/src/Native/Utils.js#L165-L179

@markhamburg, were you able to try the code out on your example?

[saschatimme]

This looks really good @edennis! I think this is ready to get merged if @markhamburg can verify that this resolves his problem.

[edennis]

@saschatimme I'll see if I can put together a simple example app that will show the "before" and "after" behavior.

[edennis]

@saschatimme I've pushed a small example app that demonstrates the broken behavior in the current version. If you run it as is you'll see the error I've described in the README. It's a bit of a pain to test my PR since you'll have to clone it locally. What I did was more or less:

cd assets/elm-stuff/packages/saschatimme/elm-phoenix
mv 0.3.1 0.3.1-old
ln -s location-of-my-pr 0.3.1

and then recompile and run again. Hope this helps convince you.

[saschatimme]

awesome work @edennis! :)

[saschatimme]

Do you have other changes planned or shall I cut an release?

[edennis]

@saschatimme Thanks! I haven't actually popped everything from the stack yet - there is one more coming, but I can't make any promises about ETA. I'd probably go ahead with a version bump now for the sake of granularity just in case anyone reports regressions.

[markhamburg]

I pulled the code into my project and tried to recreate the situation we had before and it all worked correctly this time. The diff against my own patches also confirmed that this was doing what I expected.