Behaviour When Mixing Dispatch and Dispatch Sync about citrus

[roman01la]

Thanks for reporting this. I’m on vacation now, will take a look later.

[DjebbZ]

I've been bitten by the same root cause of this problem just today.

For dispatch!, the side effects are always executed before changing the state. Here's the code for latest Citrus 3.2.
For dispatch-sync!, the execution order between the :state effect and other side effects depends on the order of traversal of Citrus's custom doseq macro. See the code.

Makes me think that there's a problem of predictability and/or coherence, or at least a documentation problem. If code needs to be changed, I don't know how to make it non-breaking.

[martinklepsch]

I think with the the recent :citrus/handler refactorings (#50 & #59) the behavior should be more consistent:

citrus/src/citrus/reconciler.cljs

Lines 104 to 114 in 8ed6d0e

	(schedule-update!
	batched-updates
	scheduled?
	(fn batch-runner []
	(let [events @queue]
	(clear-queue! queue)
	(citrus-handler this events)))))
	(dispatch-sync! [this cname event args]
	(assert (some? event) (str "dispatch! was called without event name:" (pr-str [cname event args])))
	(citrus-handler this [[cname event args]]))

Specifically :state will always be executed last:

citrus/src/citrus/reconciler.cljs

Lines 46 to 54 in 8ed6d0e

	(m/doseq [effect (dissoc effects :state)]
	(let [[eff-type effect] effect]
	(when (s/check-asserts?)
	(when-let [spec (s/get-spec eff-type)]
	(s/assert spec effect)))
	(when-let [handler (get effect-handlers eff-type)]
	(handler reconciler ctrl effect))))
	(if (contains? effects :state)
	(recur (assoc state ctrl (:state effects)) events)

I guess this is a breaking change in some ways and we will document this in the release notes.