A few notes about the current state of the proposal: defn is not sufficient, fn itself would need to be support too. Example uses defn only.
Your generator examples only show yield in one direction. This is already pretty much possible with seqs and you can write seq->gen + gen->seq functions pretty easily for those. No work needs to be done for that. The part that is currently possible via seqs is "var x = yield;" since generators/yield are two-way. Closure is changing how lanuage-in/out work. Output will be set as a feature-set specifying exactly which features should be compiled down or kept. At least that is what a recent commit [1] suggests. Input already pretty much detects the feature set when parsing, it is mostly used an optimization to skip certain work before parsing. I expect some changes here as well. There would need to be a way for processing the CLJS generated code in :none. Since no transpilation is done for those usually (since its ES3) that would be an entirely new thing. Given that most developers will be running modern environments this might not be too critical overall though. [1] https://github.com/google/closure-compiler/commit/7d00895022cb0704edd8beeabeabdcb065c365c5 On Friday, May 25, 2018 at 10:59:10 PM UTC+2, Shaun LeBron wrote: > > agreed that core.async is not relevant to this discussion. > > I suppose the main question is—how should ES6+ (ES Next) features be made > available to CLJS users. Is the story going to be "All of ES5 is > accessible from CLJS—but you must use an external JS file for ES Next > features"? > > (I added your transpile link to macros section. Also, I'd love to see > emitting ES Modules in the future, though I'm not sure of implications > there) > > On Friday, May 25, 2018 at 11:22:26 AM UTC-5, Thomas Heller wrote: >> >> Yes, core.async is not great for this but it was also not meant for this. >> The big abstraction there are channels or CSP. There are "zero" channels >> involved in your example so you could probably find a better abstraction to >> fit here. >> >> There are several different concurrency models and all of them have >> drawbacks. As I said before I'm for adding support for async/await and >> generators but core.async should not even be part of the argument IMHO. I'm >> not trying to convince anyone that core.async is the best solution ever. I >> remember vividly how desperately I wanted a proper Actor-like concurrency >> primitive when I came over from Erlang. I actually only understood the >> drawbacks of Actors after I learned about CSP. >> >> I made this example a while ago showing what Closure does when rewriting >> async/await to ES3. >> https://gist.github.com/thheller/82ca9ef8a04b58aafa4624edbac383ac >> >> We could emit that JS code directly using a few macros. The only reason >> emitting the ES6+ code makes sense is due to tool support and maybe having >> slightly more compact code, which is probably not a factor after :advanced >> is done with it. We can already interop with all of it very easily, just >> writing equivalent code is not ideal. >> >> To be honest just adding support for async/await and generators alone >> doesn't make much sense to me. At that point we should probably look into >> emitting ES6+ESM directly instead and make proper use of all the features >> that has. Given that Closure support for ES6 is getting much better that >> might make sense at some point. >> >> >> On Friday, May 25, 2018 at 4:45:22 PM UTC+2, Shaun LeBron wrote: >>> >>> Right! The proposal mentions that go-blocks must check for a closed >>> channel at every step in order to exit early. So I'll revise the >>> title—core.async cannot stop *arbitrary* go-blocks. >>> >>> For example, with this staggered animation code, it's not immediately >>> clear to me how to exit after any step. In JS, it pulls the plug for free. >>> >>> (go >>> (dotimes [_ 3] >>> (swap! game assoc :board cleared) >>> (<! (timeout 170)) >>> (swap! game assoc :board board) >>> (<! (timeout 170))) >>> (swap! game assoc :board cleared) >>> (<! (timeout 220)) >>> (swap! game assoc :board collapsed)) >>> >>> >>> Anyone coming from the Unity game engine will recognize this feature as >>> the StopCoroutine function. Core.async does not have such a feature. the >>> ergonomics matter! >>> >>> On Friday, May 25, 2018 at 3:28:49 AM UTC-5, Thomas Heller wrote: >>>> >>>> >>>>> not quite! core.async doesn't allow you to cancel a go-block (to my >>>>> knowledge), which JS allows. I added a section on this: >>>>> >>>>> >>>>> https://beta.observablehq.com/@shaunlebron/proposal-generators-and-async-functions-in-clojurescript#coreasync >>>>> >>>>> >>>> This is incorrect. Closing a channel can be used to "end" a loop. In >>>> addition it is possible to use alt! or alts! with an additional "control" >>>> channel (or more) and "selecting" which channel to work on. >>>> >>>> >>>> >>>> (defn foo [] >>>> (let [ch (async/chan)] >>>> (go (loop [i 100] >>>> (when (pos? i) >>>> (<! (async/timeout 1000)) >>>> (when (>! ch i) >>>> (recur (dec i))))) >>>> (prn :loop-terminated)) >>>> ch)) >>>> >>>> (go (let [ch (foo)] >>>> (prn (<! ch)) >>>> (prn (<! ch)) >>>> (async/close! ch) >>>> (prn (<! ch)))) >>>> >>>> >>>> >>>> In addition the "loop" is dependent on the "ch" reference. If that gets >>>> garbage collected the loop will be as well, similar to generators or >>>> async/await. >>>> >>> -- Note that posts from new members are moderated - please be patient with your first post. --- You received this message because you are subscribed to the Google Groups "ClojureScript" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojurescript+unsubscr...@googlegroups.com. To post to this group, send email to clojurescript@googlegroups.com. Visit this group at https://groups.google.com/group/clojurescript.
