Ember Fastboot
ember-fastboot.com
ember-fastboot.com
(P.S. One thing I think that's pretty meta-cool about the FastBoot site is that it is, itself, a FastBoot site, running on Heroku.)
Normally Ember app would render everything within the browser after all JS is downloaded and components etc. are processed. With Fastboot, somehow an additional, non-interfering layer of computation on the server does the same loop (without having to download anything and without significantly slowing down overall client app JS initialization) and sends the output to the client. It does it for every route separately. I'm guessing that somehow the client's processing is disabled on that first "pageview" to avoid double calculation. Is that correct? (if it is, it's very cool :)
With Fastboot, somehow an additional, non-interfering layer of computation on the server does the same loop (without having to download anything and without significantly slowing down overall client app JS initialization)
First the browser downloads the HTML, then the JavaScript, then the JavaScript runs and fetches the data via XHR. Only then does the user see the content they were after in the first place.
This actually works surprisingly well for "workspace" apps where the user is using it throughout the day (Gmail, Google Docs, etc.) On modern devices with good broadband, the difference is negligible.
But the thing this sucks for is content sites, where you aren't using an "app" but you're just clicking a link in Twitter or something. If it doesn't load within a second or so, you aren't that invested that you don't just close the tab. That has been the biggest source of pushback on frameworks like Angular and Ember for sites like this.
FastBoot bends the curve by replacing that static HTML file. Rather than serving an empty document that just points to JavaScript assets, we keep your Ember app running in Node.js on the server. When an HTTP request comes in, we direct it to Ember's router, where it figures out what models to load and components to render. When it finishes, it sends the document back to the browser.
You can think about this is as effectively outsourcing the JavaScript runtime to the server for the first load, but then the browser can take over again on subsequent navigations so it's very fast.
I think FastBoot is a great option for search crawlers, Facebook and Twitter embedding (it supports Open Graph and Twitter Cards), and supporting JavaScript-less clients. Most importantly, it's a way to get content quickly to users with a cold cache.
That said, we are planning to aggressively take advantage of App Cache and Service Worker, so ideally any second-time visitor to your site only has to fetch the raw data to see what they're after.
I'm guessing that somehow the client's processing is disabled on that first "pageview" to avoid double calculation. Is that correct? (if it is, it's very cool :)
Specifically, ember-engines provides a way to separate code into logical apps, but having them appear to be a single app. It's currently 'experimental', but in very active development.
https://github.com/dgeb/ember-engines
For us, it provides the best-of-both worlds... a single interface to the customer, without having to load up all of the code, services, etc that are used in every corner of the app.
Aren't we now just back to square one again in terms of MVC frameworks? What are the advantages here over simply implementing Django, RoR, Spring, Laravel, etc and cutting back on the JS (at least in terms of content-driven sites)?
You get all of the advantages of single page apps, without the unresponsive initial load time. There are also whole classes of apps you can't build with Django, RoR etc, so the question is a bit ridiculous IMO.
Wait, is one use-case that works out of the box to have a view that works both in text-mode browsers, like w3m and lynx (or graphical browsers with js disabled due to security or other reasons), and also "ramps up" to work well as "regular" SPAs?
Because that's one thing I've yet to see anyone really pull off: being able to define sane, simple components/widgets that also works reasonably without javascript (eg: the ability to render data as a table, and have click-able headings for sort that works both in lynx and without hitting the server in a modern js browser).
[ed: It certainly appears that http://www.ember-fastboot.com/docs/user-guide works well enough with js off, except for a couple of images not rendering.]
Btw great job!
That said, it should work just fine once the Ember app finishes loading and running in the browser. I would love it if you could file an issue on the Ember Inspector GitHub page with the error you're getting.
In addition, Glimmer 2 is near completion, and when ready will provide further speed ups to initial rendering.
Everything is working just like you would set it up with React, except that with Ember declaring data dependencies on route is existing convention. And React gives you maybe a little bit more control around that.
That makes sense, I believe that's exactly what react-async does (if I recall correctly?) with react-router.
For our project, we basically call "await ApiActions.getData(routeParams)" in the server's route, after pulling that params out, which is effectively the same idea, and it worked quite well. I'm curious how one would tackle that from a different perspective, but it seems we all end up coming to the same conclusion!
What is this? And why should I use it? I slummed through the quick start but that didn't make me any wiser.
EDIT: I read up on your blog post and you guys should definitely put some of that into the web page.
Oh come on now, lets not be hyperbolic. I'm a massive fan of Ember, but I'm a bigger fan of Redux with React; the reason that Ember can afford to be opinionated is because the team is extremely smart and takes the best winning ideas from the outside "insane" community.
The fact that Ember can distill the insanity of the community into an opinionated and usable framework (and also take wonderful ideas from things like Rails) is exactly what I mean by a source of reason.
One of ember's greatest benefits is that it adapts to change and doesn't miss out on features for very long at all (you can look at fastboot as a reaction to isomorphic react apps, or something that the ember team would have just pursued anyway). However, that benefit can also be a pitfall for newcomers to ember as it's hard to find consistent discussion, help, and resources for a framework that changes so fast.
BTW, while Ember CLI makes things much easier, it does not improve the complexity situation, it just becomes one more thing you have to learn when learning Ember (even as a newbie). What if a newbie isn't familiar with node? what if they're not sure why you're precompiling? what if they're not familiar with task runners like grunt and gulp?
Contrasted with frameworks like Angular 1, Backbone+Marionette, Ember definitely has the most rampup and complexity, not the least.
Ember CLI does file generation and a whole lot more. I'm not saying it shouldn't -- but that shouldn't be the easiest most approachable way to start with Ember.
Why should someone have to learn all the following things:
- transpiling
- nodejs
- npm & packaging
- bower
- broccoli/task runners
- livereload
... just to START with a web framework?
Maybe don't market Ember to beginners? Again, I like Ember, I think it's the most viable large framework out there right now -- but this is certainly an issue
Angular doesn't recommend any tooling, pretty much (just include the framework code and you're off to the races, write some html write some js)...
And to be fair, the same thing has been said about React, though both Ember and React allow you to use a CDN link like in the good old days.
I started a new job and went from 0 JS (other than some JQuery and knowing the syntax) to 100 with Ember. Ember is really really good for beginners, grandparent talks of having to know transpiling, broccoli/task runners and livereload but doesn't understand that you need to know none of that to get working with Ember.
Write your app by editing the files ember-cli produces. No transpiling, or add transpiling with a single 'ember install' command. Who cares about broccoli, I just edit my ember-cli-build.js file with some paths and it all works. Livereload is hardly difficult to understand, with Ember you just run "ember serve" and it also all just works.
My point though, was that after you build on all this complexity that you don't understand (and don't have to deal with), when something goes wrong, you're in for a world of hurt. But maybe that's not an issue
As a developer, one becomes productive when one realize when to put the blinders on, and when to take them off. As such I for one, love that I don't need knowledge of x86 assembler, chip design, or signal processing until the problem at hand actually requires them.
Ember-cli aspires to keep developers focused on features, not orthogonal tech. That is unless they need to peel back that layer of the onion, and dive in. Even then, the goal is for only a few community members to dive in, explore the problem space, and ultimately contribute the solution. Next release, all community members benefit, without also having to invest (until the point where they have a specific itch to scratch).
Abstractions hurt when they leak, as such we must aspire to provide the best abstractions we can (at each layer), and this is only possible in collaboration with an eager and enthusiastic community.
An symptom of a curated solution, is all aspects of the stack evolve to work together. Mitigate abstraction leaks at the various boundaries.
Also react's documentation is terribad. I'm in the middle of teaching someone who is completely new to JS how to use React, and the documentation has been horrendous (for newbies), reasons:
- JSX is super hard to properly understand if you only know HTML and JS and don't know they normally don't mix, and aren't actually mixing
- Requiring precompilation (no more JSXTransformer, no more babel-browser)
- The entire concept of components-all-the-way-down
- Documentation (Reference) section that does not differentiate between code blocks you can use and function signatures
- Hints on how to deal with routing
- No explanation of where to store data/state
I, as an experienced developer, can very easily understand React (I've used it on 2 projects now I think, and I like it), but there is so much for beginners to stumble on.
You can very well use React/Angular(1, atleast)/etc. just by including a script file in your markup and continuing to write your application code. There was a tutorial series for React posted on HN sometime before that made you learn React just this way.
This is probably the thing I hate the most from client-side apps. Can't wait to test this and Angular 2 on production!
Famous last words :)
[1]: https://github.com/mhart/react-server-example
[2]: https://angular.io
There is a huge gulf between "synchronously render a component in Node" and "asynchronously boot an app, marshall async data, render an async UI, and do it concurrently."
I touch on this a little bit in this talk[1], but the bulk of the work we've done over the last year is conventionalizing app boot and figuring out how to run multiple instances concurrently, cheaply.
Angular 2 is much more similar to what we're doing with FastBoot, and I'm excited to see where they go with that and crosspollinate ideas. (The work they're doing with Web Worker rendering is also very exciting.)
I am a software engineer who usually develops also for the client side and I follow and experiment with the JS trends because I find them interesting. However, I don't understand what you mean by these. Could you please be more specific? Thank you.
It looks really cool... if I'm understanding it correctly, it can hot-load code in production, which is awesome and scary.
It appears to be a quickstart, so what is upgrading like?
It might be easier to get server-side rendering working with React, but pretending that it's not the same thing is just condescending. There are dozens of examples out there showing how to render React pages on the server, including fetching the necessary data for that rendering.
The mechanisms you use might be different, but pretending they're not analogous is duplicitous at best. Both let you ship a fully rendered first view to the client.
It is fundamentally different, as I outlined above. That's not downplaying the work of React—they've done significant work in advancing the state of the art of DOM rendering that has hugely inspired the work of both Glimmer and Glimmer 2, as we have said frequently and publicly.
I have tremendous respect for the Angular and React teams and consider many contributors to both to be personal friends. I'm not downplaying their work—it just really, really, really is not the same thing.
That said, everything you're describing – "asynchronously boot an app, marshall async data, render an async UI, and do it concurrently" – is really, really, really the same thing we're (my team at least) doing with React. People aren't just using `renderToString` and calling it a day.
Is there an 'analagous initiative' in React? No.
Can you end up with the same result in React? Yes.
There are people doing this in React, but each team needs to do it slightly differently because their app is setup differently. There are guides and examples on how to do it, but unless you setup your project exactly like the original author, you'll have to figure out portions of it on your own.
On top of that, you won't know if the modules that you are using will play well with the offline rendering.
The convention-driven approach IS the product... You make it sound like that is just a little bit of polish, maybe a page or two of documentation... But it took over a year of planning and architecture to make all of the changes. (There was a great demo of fastboot a year ago) Because of the planning and framework changes, however, every addon that follows the conventions will 'just work' with fastboot.
When Tom says "An App"... he means "Any App"... and that isn't something that I've seen being worked towards by the React community. It IS being done by other communities, as seen in this thread, so I think that it's valid for Tom to call this out as not being analogous.
The methods might very well be different, but the result is the same.
Pretending that you can't do server-side rendering with React is not helpful and untrue. Just because they haven't put the same work into conventionalizing it does not mean that analogous initiatives don't exist for React.
There are plenty of guides and tutorials out there showing how to do server side rendering in React.
You're right that tomdale came off condescending (as he often does when promoting/defending ember...ah well). But it certainly isn't "duplicitous at best"!
Analogous doesn't mean identical. The results are similar, even though what's happening under the hood is different.
> do it concurrently
Does ember renders concurrently?
> There is a huge gulf between "synchronously render a component in Node" and "asynchronously boot an app, marshall async data, render an async UI, and do it concurrently."
I dont think it is fair to say it that way. React implementation doing all this (except async/streaming rendering) is man-week at worst in existing medium sized project, and man-day for a new project with no pre-existing code.
The goal of those small examples is to illustrate that you don't need complex frameworks to achieve server-side rendering with data fetching and routing.
I have no doubt that FastBoot was a huge undertaking, and I think a large part of the reason for that is that it's aiming to fit into a very comprehensive existing framework, with pieces for routing, fetching data, "application instances", etc – there's a lot more work involved in that (for the framework author I mean, not the end user) than there is starting off with just a view library and adding your own pieces around that.
(In the comment above Tom is suggesting fastboot goes significantly further than this, to make your 'client-side app' render server-side).
you write a client-side app and can render on server via Node and hydrate/attach the js on client after the dumped html. you get "instant" initial render and "progressive enhancement" once the js executes.
...or do everything on client.
a more fleshed-out demo would be valuable for a more realistic comparison. they mention the non-congruence between ajax vs node's fetch, but this is easy to shim so the same code works on both.
the main complexity is the view rendering and post-render hydration.
* Fully asynchronous rendering, so you don't have to awkwardly architect your app so that it can be rendered synchronously.
* Everything is fully progressively loaded. This means if you go to a particular page in your app, only that page's JavaScript and CSS will be downloaded in the client. Additionally the correct css link elements will be inserted.
* Caching XHR requests so that they are not repeated on the client (data used to render is included in the page).
What makes our solution unique is that you have to think about the server very little, if at all. If you need to make a request to services, just make it in your code. No additional wiring is needed and everything will be server rendered.
Much of this is possible because of Zones, a spec that is being worked on for standardization in TC39. We have a library that implements Zones[2] with SSR in mind. This is what makes XHR caching possible, for example. Check out this simple jQuery example app[3] (using jsdom on the server) to see how easy Zones make things.
I did a talk at Node Interactive this year about SSR and what goes into a good SSR solution: https://www.youtube.com/watch?v=wRYdrfrL6ZQ
[1]https://donejs.com/ [2]https://github.com/canjs/can-zone [3]https://github.com/canjs/can-zone-jquery-example
What is an example of something a dev might have to think about with fastboot that they don't have to think about with DoneJS?
Having access to .cljc (clojure files that can be loaded by either CLJS or standard CLJ) helps tremendously here. This means the server-side happens in Clojure on the JVM, so all of the async worries just go away entirely.
[0] - http://markojs.com/
For what its worth, FastRender was first released in 2013. The just-released Meteor 1.3 now has ES2016 module support which should enable more improvements in this area.
1. https://meteorhacks.com/fast-render-internals-and-how-it-wor...
Build error
The Broccoli Plugin: [object Object] failed with:
RangeError: Maximum call stack size exceeded
at new Error (native)
at Error (native)
at Object.fs.mkdirSync (fs.js:794:18)
at sync (C:\Users\marke\ember\github-fastboot- example\node_modules\ember- network\node_modules\broccoli- templater\node_modules\broccoli-stew\node_modules\broccoli- funnel\node_modules\mkdirp\index.js:71:13)
at sync (C:\Users\marke\ember\github-fastboot- example\node_modules\ember- network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:77:24)
at sync (C:\Users\marke\ember\github-fastboot-example\node_modules\ember-network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:78:17)
at sync (C:\Users\marke\ember\github-fastboot-example\node_modules\ember-network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:78:17)
at sync (C:\Users\marke\ember\github-fastboot-example\node_modules\ember-network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:78:17)
at sync (C:\Users\marke\ember\github-fastboot-example\node_modules\ember-network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:78:17)
at sync (C:\Users\marke\ember\github-fastboot-example\node_modules\ember-network\node_modules\broccoli-templater\node_modules\broccoli-stew\node_modules\broccoli-funnel\node_modules\mkdirp\index.js:78:17)
The broccoli plugin was instantiated at:
undefined
Looking around on the site I didn't see a link to report something like this, and I didn't immediately think to goto github with it.
Please excuse my ignorance.
In my eyes Ember has become an old-school Rails like Blob and newer frameworks are more about modules and "Bring Your Own Tools". I mean who, besides Ember, uses Broccoli?!
* Code generators can speed up development * They give newbies a better idea of how the framework should be treated * They give the framework a more dependable layout * They still provide the escape hatch of not using them!
What would a dev have failed at simply because they built a code generator?
Creating an understandable API and making stuff easily extensible.
I often have the feeling these code generators try to hide bad design decisions, which resulted in a huge amount of boilerplate code, that wouldn't be needed if things were designed different.
And you can't even blame most frameworks for it. Ember has so much history that you can't simply throw every thing out, because "now animations or server-rendering is a PITA", so devs maneuver around it with code generators.
There are valid reasons to have them. But when I have to choose frameworks for a new project, code generators are definitely red flags for me.
I don't have much of a problem with this controller/route/model generation stuff, that you mentioned. I wouldn't use it, but I can see why people save a few minutes.
I don't see how you have a large ecosystem that is still friendly to a new developer without code generators unless you want huge boilerplate. A frontend requires a lot of moving parts, and letting someone else take care of that for me is important.