[Fix clojure-emacs/cider#1320] Do state tracking in a future

[Malabarba]

This should fix all lags associated with the new namespace tracking,
both the long delay at first connection and the ~1sec lag in the REPL.

[bbatsov]

Won't an agent be a better fit for this?

[Malabarba]

Is it ok if the function used to update agents in not pure?

[expez]

I think you're risking retries when using agents (but we only use the agent from one thread), but if the only IO is reading from disk then that should be fine.

[bbatsov]

I think agents never trigger retries, so this should not be a concern. After all people are implementing log systems and things like this using agents. In a transaction an agent won't be executed until the transaction commits, but this is different.

[Malabarba]

I'm running some tests here, and it looks like agents do not trigger retries indeed.

(let [at (agent 0)
      c (atom 0)]
  (dotimes [x 1000]
    (future (Thread/sleep 200)
            (send at #(do (swap! c inc)
                          (inc %)))))
  (Thread/sleep 2000)
  (println @c))

This always prints 1000. Whereas, if I replace (agent 0) with (atom 0) and replace send with swap! then I always get at least 1030.

Not a super rigorous test, but it's convinced me. :)

[Malabarba]

https://clojurians.slack.com/archives/clojure/p1442501641001191

[expez]

I'm not on slack but I just looked at the implementation and you guys are indeed right. There are no retries because the agent holds an ActionQueue internally.

[bbatsov]

So, it seems we have clarity here. Agent it is! :-)

[bbatsov]

Apart from the failing tests the code looks good to me.