Ember 2.0 has gone great lengths towards making everything a component. With routable components coming soon, controllers will probably phased out as well.
Context
However, there is a recurring problem I face when building User interface, which I don't have a satisfying pattern for so far: user interface state.
What am I taking about?
Selection state
Current focus
Folded/unfolded state in some tree display
Basically, any state that is not part of the actual data, yet has to be tracked on an object-by-object basis. In the past, this used to be done with Controllers, acting as proxies to models. This approach is now obsolete. The new approach is Components everywhere, for the better. Component does the bookkeeping, tracks transient state and you get actions back.
A typical pattern I have, though, is with shared state, such as a list with selectable items.
The issue
Building a list component, with the following requirements:
Each item can be selected.
Selection state changes the DOM (some class bindings).
User can draw a rectangle in the list component to select several items at once.
Ideally, the whole behavior can be abstracted out as a mixin.
Question: where does the selection flag live?
Attempts
1) Everything is a component.
I make each item a sub-component, say {{my-list-item}}. The component tracks the selection state. Problem: how can the list component update the selection state?
2) Move state out of sub-components.
Put it on the list component. Within a separate state array alongside the item list. Pros: the list has all the state it needs. Cons: it's a nightmare to keep it synced when items are added or removed from the list. And it means the sub-component have no access to the state. Or maybe I could pass it to them as a bound value?
3) Reintroducing proxies.
Coming to think of it, there is a way to share some state: put it on the models. Well, not on the actual models so as not to pollute them with local state, but by setting up an ArrayProxy that will return some ObjectProxy for every item, to hold the state.
Pros: this is the only solution I managed to implement completely. Cons: encapsulation and decapsulation of items is a hassle. Also, after a few layers of being passed around, get and set have to go through 4 ou 5 proxies, which I fear will be a problem with performance.
Also, it does not work well for mixins. Should I want to abstract out some HasSelection mixin, and a HasFoldableItems mixin, and a Sortable mixin, they all need some state.
Back to the drawing board
Is there some better pattern I have not found?
I found the following relevant questions, but that led me nowhere:
Ember.js - where should interface state be stored? (2012, suggests something close to my #3 above)
Road to Ember 2.0 - High level Ember app structure feedback? (some of the key questions in the list are relevant to this one)
Great question - I actually went to one of the core ember team members to find out and the answer currently is services. Because the component is best left stateless (for the most part) you can leverage a service to persist this state that doesn't fit into a server persisted model.
Here is a great example that Stef Penner put together showing how you might save a "email draft" in your ember app (that isn't persisted backend)
https://github.com/stefanpenner/ember-state-services
Example component for reference (from the github project above)
export default Ember.Component.extend({
tagName: 'form',
editEmailService: Ember.inject.service('email-edit'),
state: Ember.computed('email', function() {
return this.editEmailService.stateFor(this.get('email'));
}).readOnly(),
actions: {
save() {
this.get('state').applyChanges();
this.sendAction('on-save', this.get('email'));
},
cancel() {
this.get('state').discardChanges();
this.sendAction('on-cancel', this.get('email'));
}
}
});
Related
I am trying to build a UI with left side bar having filters and right side having actual filtered data.
For loading data into the dynamic part of the UI(right side), which approach is considered better in terms of code quality and app performance ?
Use sub routes (for dynamic part of the UI)
Use separate components that load their own data (for dynamic part of
the UI)
There is not a direct correct answer for that; you can use both ways but here is a few things to consider and in the end I generally prefer to use sub-routes due to the following:
Waiting for UI to load: In case you are using separate components to load their own data; then you need to handle the loading state of the components. What I mean is; if you simply use sub-routes; then model hooks (model, beforeModel, etc.) will wait for the promises to be solved before displaying the data. If you simply provide a loading template (see the guide for details) it will be displayed by default. In case you use components, you might need to deal with displaying an overlay/spinner to give a better UX.
Error Handling: Similarly like loading state management; Ember has already built in support for error handling during route hook methods. You will need to deal with that on your own if you prefer components to make the remote calls. (See guide for details)
Application State: Ember is SPA framework; it is common practice to synchronize application state with the URL. If you use sub-routes; you can simply make use of the query parameters (see the guide for details) and you will be able to share the URL with others and the application will load with the same state. It is a little bit trickier to do the same with components; you still need to use query parameters within the routes and pass the parameters to the components to do that.
Use of component hook methods: If you intend to use the components then you will most likely need to use component hook methods to open the application with default filter values. This means you will need to make some remote call to the server within one or more of init, willRender, didReceiveAttrs component hook methods. I personally do not like remote calling within those methods; because I feel like this should better be done within routes and data should be passed to the components; but this is my personal taste of coding that you should approach the case differently and this is perfectly fine.
Data down, actions up keeps components flexible
In your specific example, I'll propose a third option: separate components that emit actions, have their data loaded by the route's controller, and never manipulate their passed parameters directly in alignment with DDAU.
I would have one component, search-filter searchParams=searchParams onFilterChange=(action 'filterChanged'), for the search filter and another component that is search-results data=searchResults to display the data.
Let's look at the search filter first. Using actions provides you with maximum flexibility since the search filter simply emits some sort of search object on change. Your controller action would look like:
actions: {
filterChanged(searchParams){
this.set('searchParams', searchParams);
//make the search and then update `searchResults`
}
}
which means your search-filter component would aggregate all of the search filter fields into a single search object that's used as the lone parameter of the onFilterChange.
You may think now, "well, why not just do the searching from within the component?" You can, but doing so in a DRY way would mean that on load, you first pass params to the component, a default search is made on didInsertElement which emits a result in an action, which updates the searchResults value. I find this control flow to not be the most obvious. Furthermore, you'd probably need to emit an onSearchError callback, and then potentially other actions / helper options if the act of searching / what search filter params can be applied together ever becomes conditionally dependent on the page in the app.
A component that takes in a search object and emits an action every time a search filter field changes is dead simple to reason about. Since the searchParams are one-way bound, any route that uses this component in it's template can control whether a field field updates (by optionally preventing the updating of searchParams in an invalid case) or whether the search ever fires based of validation rules that may differ between routes. Plus, theres no mocking of dependencies during unit testing.
Think twice before using subroutes
For the subroutes part of your question, I've found deeply nested routes to almost always be an antipattern. By deeply, I mean beyond app->first-child->second child where the first child is a sort of menu like structure that controls the changing between the different displays at the second child level by simple {{link-to}} helpers. If I have to share state between parents and children, I create a first-child-routes-shared-state service rather than doing the modelFor or controllerFor song and dance.
You must also consider when debating using children route vs handlebars {{if}} {{else}} sections whether the back button behavior should return to the previous step or return to the route before you entered the whole section. In a Wire transfer wizard that goes from create -> review -> complete, should I really be able to press the back button from complete to review after already having made the payment?
In the searchFilter + displayData case, they're always in the same route for me. If the search values need to be persistent on URL refresh, I opt for query params.
Lastly, note well that just because /users/:id/profile implies nesting, you can also just use this.route('user-profile', { 'path' : 'users/:id/profile' }) and avoid the nesting altogether.
I know this question might seem a little duplicate but the other version of this question is old and some of the content (such as Views) aren't even a part of ember anymore.
I'm about 4 weeks into my internship as a front-end developer working with EmberJS. I still don't understand when it's better to use the route over the controller or vice-versa. It seems to me that every action in the route can also be used in the controller.
The one recent thing I heard was that ember routes should be stateless where as controllers should be stateful.
What is the current state of controllers and routes. When should one be used over the other?
Consider the following example to understand the state of a controller (or route, or anything), in simple terms and in current context -- lets say you have a page (like a form) with three tabs; each tab can be considered as a state - it would call different components based on the state (or the tab you are in). Now if you would happen to go back for some reason, and hit the form link again, you would see that the state would remain the same. (if you were on tab 2 when you hit back, on returning to the form, you would still be on tab 2).
So to maintain these states, controllers are the way to go, since they are singletons. Route would have lost that information, and started fresh. So basically your variables/objects in a controller would define the 'state'.
Route-actions can be as easily used as controller actions- see https://github.com/DockYard/ember-route-action-helper. So if your template for this route is just using model as the object directly, and you don't need to maintain the 'state', you can pretty much do without your controller.
But if your template was using variables which needed manipulation, you would need controller.
Hope this helps!
I have an application scheme implemented in Ember that basically follows this layout:
The concept is that the user can interact with features on the Map View (which is always present) and basic navigation occurs between various views in the Parent View and an arbitrary stack of Sub Views. The user can create new features on the map, edit existing ones etc.
The URL for a particular feature could be /features/123/edit
Since input into my interaction panels is very much dependent on interaction with the Map View (drawing a polygon, placing a marker etc.) my controllers for these views are setup to "need" the Map Controller. When a particular panel view is present interaction with the map should affect the panels in various ways.
My question is - how does one scale such tight controller coupling? I essentially need to switch between different Map modes based on which panel is currently active. I also, I believe, need to observe events on the map and act upon such events depending on the current active panel.
I setup a proof of concept where a certain Sub View Controller observes certain properties of the Map Controller (with for example .observes("controllers.map.activecoords") however, such an observer will continue to trigger even after the user has navigated away from the particular Sub view (i.e. as soon as the controller has been initialised). Must I setup and tear down such observers manually (i.e. using addObserver) when entering and leaving the route? Is this the right pattern? I've gotten the impression that requires that I manually remove all such observers during transitions to avoid unexpected behaviour and memory leaks.
Perhaps I'm going about this completely the wrong way? Are there any others patterns that fits my use case with an always present map with different states and intercommunication with interaction panels?
Perhaps the architecture of your application shouldn't be connecting controllers, but rather ask yourself "What is the model here, really?"
In each case, I think the model is your map, or at least its "contents". The features are really decorating the interaction with your central model, which is the map, underlying it all.
So really you have a single model here. You effectively have a map resource, and many feature routes on that resource, viewed from the URL/API.
The question is now not so much "how to manage a dependency hierarchy between controllers?" as "how do I manage a view and a subview on the same model?" which is answered quite simply in the standard ember nested route. Your outer view is always present and it has an outlet which is where your feature goes. How it's rendered is the inverse of how the standard nested route is rendered, but nevertheless it's the same pattern.
So, the TL;DR answer is... go through your model and the routes... use them to talk about your common data: the model is, afterall, your data, and the controllers and views are simply augmenting and enabling the user experience, presentation and interaction of them.
Most of these sorts of architectural problems can be resolved by moving the data up and down the hierarchy (template view / component / controller / route / model) until you find a place which is low enough to still be accessible to objects that need access to it, yet high enough that it makes sense by not being tightly bound with too many things, but still is in the right spot and "feels right".
If it's too high, you'll tend to be doing the wrong kind of work with your objects (ie controllers should not be mucking with view mechanics, models shouldn't really have presentation stuff in them, etc.
If it's too low, you'll tend to be doing too much work with the framework and you'll tend to be repeating yourself a lot... ie controllers will be doing a lot of work to get the data they need, for example.
I'm been playing with ember.js off and on the last few months and always end up asking "what role does the controller play" because it usually looks something like this when I'm done
App.Controller = Ember.Controller.extend({
content: null
});
If I have any basic validation requirements I tend to do these in the view itself.
If I need to populate a controllers content I do this in the router (though I'd prefer to reduce the # of responsibilities this object seems to have in the current version of ember).
My models are persisted using a separate adapter via ember-data with some domain logic like behavior when needed.
What I'm left with in every app I've built is a thin (very very logic less) controller as shown above. I do find this is a good place to do sorting / pagination / search / filter / etc on the items but what else should this controller object be responsible for in a well design ember.js app?
As you say, sorting / pagination / search / filter (BTW most of those is supported by Ember.ArrayController)
I think other stuff could be computed properties around the content (model), and then the view relies on those computed properties instead of doing the job by itself.
Perhaps it could be a good place to track and persist a view state (for example collapse/expand), which cannot be persisted in the view itself because it's destroyed/created at each time.
I seem to be getting more and more confounded at what appears, on the surface at least to be pretty basic architectural questions regarding building ember apps.
Which Controller type?
In the last month or so, I've seen people implement controllers through Ember.Controller, Ember.ArrayController, Ember.ObjectController, and Ember.ArrayProxy. Removing ArrayController and ArrayProxy (due to them being identical), what are common use cases between each type?
So far, i've been able to gather that:
ArrayControllers/Proxies should be used when you have n elements within the view you intend to control
ObjectControllers should be used when the view is simple enough to maintain it's state in a single object, or be a single instance of a model's object.
Controllers --- ? No idea.
What are some basic differences between the controller types? There doesn't seem to be concrete information on when to use which, and for which use case. The API docs are good at telling me the nitty gritty of each of them, but not WHEN to use each.
The relationship between a View and a Controller can be baffling
When a View is connected via a routes ConnectOutlets function call, what exactly happens between the controller and the view?
Are events tied into the view itself (which seems to be the case) and if so, where on earth do you interact with the controller singleton to perform CRUD-esq things on its properties? this.get('controllerName') doesn't seem to do the trick, and nearly each post or tutorial or code sample out there does this a different way.
Models that aren't
I realize that Ember Data looks to help solve some of the more irritating parts of dealing with data and keeping it in sync, but at a larger perspective, in the concept of "MVC", ember doesn't really seem to have a Model of any kind. It's just some object that gets subclassed from a specific thing and then tracked....somewhere? Somehow? Magical?
#trek sufficed that an Ember.Object could manage ajax'ing data and handling state on the client just fine, but if you look at something like the todomvc.com ember app, it uses a localStorage paradigm that is COMPLETELY different in implementation then everything i've looked at.
How EXACTLY should the 'Model' part of the MVC equation be done here?
Views make me want to murder children
There seems to be a significant number of ways to construct a "view" in terms of displaying markup to a user.
ContainerViews, using subviews / childviews
nested outlets
Handlebars templates + an outlet
using #each foo in controller
Injection through literals (template: Ember.Handlebars.compile('<h1>foo</h1>') etc)
With that in mind, what's the 'proper' way to build modular UI components with ember? This more than anything is a major pain point for the adoption of this framework for me.
I love the direction that Ember is going with application development on the web. The concepts seem simple, but the verbosity is that of Objective-C (which makes sense given it's lineage) but I swear to god I feel like i'm fighting the god damned framework more than i'm actually working on my application. The verbosity of the syntax and the lack of structured documentation outside of API documentation (which lets face it, 300k of javascript is a significant amount of code to throw some breakpoints down and try to debug your issues).
I realize the challenge that you guys are up against, but hopefully this at least makes you pause for a minute and think of how you could make life easier for the incoming developer who's worked with other frameworks (or hell, even worked within an MVC framework, like rails or django or backbone or angular) and say "this is how we think ember should be used".
Take some of the opinionated software design decisions and apply them toward the community. We'll do nothing but be cheerleaders for you if you do it, promise.
Please don't hurt any children. AFAIK the ember-core team are all over 18, so any ember-view-related frustration is clearly better directed towards adults. With that in mind...
Which Controller Type?
You've got the "what" right, but maybe missing the "why". Controller can be a little misleading, especially coming from rails. Think of these controller singletons as representing the state (in-memory) of your application.
Which kind of controller to use depends on what is required for that part of your application. For example, a back-of-napkin sketch for any app might have a topnav, postList, postDetails section. In ember, each is represented by one or more view/controller pairs. In this app I would expect to see ApplicationController and NavigationController extending Ember.Controler while postList would extend ArrayController and PostDetails would be an ObjectController.
You could do it all using just Ember.Controller but ObjectController and ArrayController are really useful for wrapping model data. Any non-trivial ember app will probably use all three.
The relationship between a View and a Controller
A controller's job is to provide the context in which the view will be rendered. Ideally you'd like to keep logic out of views, so a typical controller will have lots of computed properties to do things like:
transform data from the underlying model objects
sort/filter/select a list of objects
reflect application state
whats the deal with connectOutlets? This is where you should be using the requested route/context to decide which views/data should be plugged into the outlets of your application. The controller's connectOutlet method has a bunch of magic to make it easy, but maybe too much magic. What happens (afaik) when you call: parentcontroller.connectOutlet 'child' is
Ember creates an instance of ChildView
The {{outlet}} handlebars helper in parentController's view is bound to this childView instance
The childView is rendered with the router.childController singleton as it's context
where to do crud stuff?: Typically in an action on the router. This seems crazy at first. Think of ember router not like rails but as a stateManager that just happens to also handle routing. In near-future router API will change to make this more clear. Anyway, use router actions to do things like create model instances, commit/rollback transactions and trigger state change. This is easy to do if you use the handlebars {{action}} helper for links/buttons as it targets the router by default.
Views on the other hand should have logic for "reacting to browser events" - that means really low-level stuff like show/hide something on mouseover or integrate with 3rd party libraries to do effects and animations.
You might find this screencast helpful in understanding how to do CRUD-esq things:
http://blog.bigbinary.com/2012/09/06/crud-application-in-emberjs.html
Models WTF?
Agreed in Ember any object could be used as a 'Model'. I think #trek does a good job of demonstrating how one might accomplish this via Ember.Object. This works great for a simple app, and six months back maybe would've been your best bet as ember-data was really immature. I'm not clear on the history of ember's todomvc app, but for sure it was written months ago. For sure it should be updated, but meantime I'd not recommend using it to learn about current ember best-practices.
Instead, you should go with ember-data. Over the last few months it has really evolved and should be the default choice for any new, non-trivial ember app. #tomdale just gave a great presentation on this topic, I'd recommend having a look: https://speakerdeck.com/tomdale/ember-data-internals
what's the 'proper' way to build modular UI components with ember?
For building modular UI components:
ContainerViews, using subviews / childviews
Injection through literals (template: Ember.Handlebars.compile ...)
For building an individual application:
nested outlets
Handlebars templates + an outlet
using #each foo in controller
Building modular UI components is a totally different problem than building an application. Ember.View and it's subclasses were designed for this purpose. You can easily extend/combine them to compose widgets with custom behaviors and share those widgets across applications.
At least that's how i've seen it done. If they are for internal use could also reference handlebars templates instead of object literals, but if planning to distribute the object literals approach seems best.
A great real-world example of this is the ember-bootstrap project. I learned a lot about working with ember-views by reading through that project's source. http://emberjs-addons.github.com/ember-bootstrap/
TLDR
Pick controller that maps to type of data being represented
Controllers provide context for the view and remember application state
Use ember data for your models
Use subclasses of Ember.View to make components
Be nice to children