Not enough frameworks make this as seamless as it should be. So it's refreshing to see it done here.
I've been looking at Phoenix (Elixir not Erlang) but the problem is that I'll have to do logic to detect when WS are available and do the fallback myself which results in more code server side.
I think this is something that should be handled by the framework itself.
> Not enough frameworks make this as seamless as it should be.
There are lots of options for Node.js, for both low-level socket-like libraries (SockJS, SocketIO, EngineIO), and frameworks (Meteor, Derby, SocketStream).
Thinking of this, it's a pity that SocketStream failed to gain traction. It has a lot, technically, and was way ahead of its time (in 2011). Beside the middleware framework for WebSockets, it also provides
* RPC and Pub/Sub, built on top of the middleware stack,
* live reload for development, including swapping CSS files without reloading the full age, which enables
* automated, continuous testing in as many browsers as you like (keep your test pages open, they are reloaded on file change)
* a Browserify-like environment to structure the client code.
* an elaborated asset pipeline, that handles JavaScript, HTML and CSS preprocessors, and packs/compresses everything for production.https://github.com/socketstream/socketstream
Both the socket layer and the asset preprocessors are based on plugins, which makes the framework flexible.
A really good system, that was plagued by a severe case of not second but third system syndrome, which scared users away, since they (rightfully) feared that the rewrites would leave them in the cold.
The original developer (Owen Barnes) is a visionay and a brillant programmer, but he still failed to execute by never being satisfied with the implementation details. He has now moved on, and the framework is being maintained by Paul Jensen. Things have now stabilized. A third rewrite of the framework has been canned. The frameworks is still as good, but it remains marginal because Meteor managed to get the mind share.
I'm the author of Phoenix. I'm glad to hear you're giving it a look! I spent a lot of time thinking about the WS layer and fallback support. Ultimately I settled for standard WS with a small multiplexing layer on top. For those that need fallback, they can drop in websocket-js (flash fallback, compatible with native WS api). Have you taken at look at this for your fallback support?
https://github.com/gimite/web-socket-js
Phoenix is still far from 1.0, so any and all feedback is welcome.
I'm new to Elixir (and FP in general), but I really like the language. I came across a couple of youtube videos on Phoenix , and it looks really nice (reminds me of Rails). I think something like Sinatra would be even nicer, for 2 reasons:
1. the standard these days is to have a rest api serving json to web/mobile clients, so a Rails-like framework seems overkill
2. Pretty much all my friends who are learning web dev on their own, find Sinatra/Flask way easier to get started with than Rails/Django.
Regardless, Elixir (and Phoenix) is a great leap forward compared to the other options for building concurrent, reliable servers. I hope more people try it out before falling for hype/marketing (cough node.js cough mongodb cough cough)
JSON doesn't really fit very well into Erlang structures (at least in r16; I haven't played with r17 yet); so I don't think not using it is that crazy.
Speaking of json in general, this reminds me of a part of a talk by Joe Armstrong he gave at the React 2014 conf (K things I know about building Resilient Reactive Sytems)
Not related to Erlang necessarily, he talks about how at Ericsson they build these cell phone network-to-internet gateways and now they go to great lengths to pack data as tightly as possible on the communication channel, bit by bit, to conserve shared bandwidth and then "here come the users and shove json across the pipe, what a travesty!" -- it was a funny statement, and I never really thought of it that way. So I guess shoving binary across websockets like N2O is not totally insane.
I never understood that argument. The reason you can have nice text based formats is that it will all be compressed by the server anyway, or am I missing something?
No. The reason people use text based formats is so that they could be read by humans without additional tools. Compression reduces the size but it's not reasonable to claim it works better then a binary format designed for light footprint.
Joe also made a funny comment on that if you listen to his talk. He said how "People say ASCII is good because you can see what's on the wire. Well can they see electrons bouncing up and down the wire? They see what a program chooses to let them see"
But even then you're left with compressed text. If you're concerned about the amount of data you're sending on the bit level, then you're throwin all that out the second you start pumping out JSON.
I think there's still room to innovate in the Erlang/Web space. Chicago Boss was a nice improvement because it ties so much different stuff together. It does have a few issues though, like not being very 'Erlangy' in places: it uses lots of parse transform magic. I think it's the right direction though, in that it's fairly general purpose.
Here are some not so conventional things going on:
* Write your page in Erlang.
* Even translates Erlang to JS as a parse transform.
* Websockets (with a fallback mechanism) is the default connection mechanism.
* Don't want to use JSON for some crazy reason? That's alright, user Bert (and ship binary encoded Erlang terms to the browser!).
* Can render stuff on the server and send the whole thing over the websocket connection.
https://github.com/5HT/n2o