The question, in brief:
In MVC, how do you distinguish between a checkbox click (or a selectbox or listbox change) from a human meaning "Controller, modify the model", and a checkbox click (or a selectbox or listbox change) from the Controller meaning "I'm updating the view because the model has changed"?
The example:
I have a JS app (all one big HTML+JS page; there's a server behind it, and AJAX going on, but it's not important to the example) which has the notion of "Vertices" connected by "Edges". The UI lets you add and remove Vertices on a map, and enable or disable Edges between pairs of Vertices.
There are two ways to disable an Edge from Vertex A to Vertex B:
click on the Edge to make the "Edge Details" window provide you with a "Disable This Edge" button; or
click on Vertex A (or B) to make the "Vertex Details" window provide you with a checklist of nearby Vertices, from which you can uncheck Vertex B (or A).
Here's how this works under the hood in MVC (but see the end of this post for an update, where I correct problems in my understanding):
Model: a list of Vertex objects and a list of Edge objects.
View: a GMaps UI, with markers and polylines, plus checkboxes and buttons and "Vertex Details" and "Edge Details" DIVs.
Controller:
JS functions that update the model when events on the checkboxes and buttons fire; and
JS functions that update the view when events on the models fire.
Here's the specific inelegance:
The user has the Vertex Details Window focused on Vertex A, and the Edge Details Window focused on the Edge from Vertex A to Vertex B.
The user clicks "Disable This Edge" in the Edge Details window.
Controller function 1 gets the click event, and calls disable() on the Edge model object.
The Edge model object fires the "I just got disabled" event.
Controller function 2 receives the "I just got disabled" event, and
redraws the Edge Details Window to say "I'm disabled!" and
unchecks Vertex B in the Vertex Details Window.
Crap! This fires Controller function 1 again, which was listening for UI events that mean an edge was disabled!
So there's an unnecessary re-update of the Model, and re-update of the View. In a more complex view with events that fire events that fire events, this can make for dozens of extraneous updates!
Update: a great answer.
I misunderstood MVC a bit. I don't have just one View, as I described above: I have several Views into several Models. In particular, I have a checkbox-list View of Edges to a particular Node, and a separate, "detailed window-style" View of an Edge.
Furthermore, I shouldn't have one controller function updating all views when the Model changes: each View should modify itself when the Model changes.
So if each View registers for "state updated" events on the Model, and each View updates itself upon receipt of those events, then the answer to my circular events question is simply this:
The checkbox-list View will disable checkbox events for the moment that it is updating the checkboxes after a Model state change.
Now if a user disables an Edge via the Edge Detail window, the Controller updates the Edge Model, the checkbox-list View receives notification of the update, and the checkbox-list View is smart enough to silence checkbox events while changing the status of the appropriate checkbox.
This is much more palatable than my original solution, where one Controller updates ALL Views -- and thus has to know which views need special care and feeding to avoid loops. Instead, only the single View with troublesome UI elements has to deal with the problem.
Thanks to those who answered my question!
Just to recap the MVC model. Views should generally update themselves. Here's how it works: a controller changes the state of the model, the model sends updates to its views, the views pull in new state from the model and update themselves. While controllers and views are generally bundled (i.e. drilling down on data in a graphic representation) they should never interact directly, only through the model. This in general of course.
So the JS functions that update your views are not actually controllers, which is an important distinction. They should be considered part of your view. This might not be helpful to the problem at hand but I thought it merited pointing out.
You can also not delete your model, I assume you mean you're deleting someting from your model, since no views or controllers can actually exist (or be in a functional state) if they're not backed by a model.
Not being a JS code jockey and not having used gmaps I don't really see where the problem is. Does changing the state of a checkbox(checked property) fire the onClick() event? It really shouldn't IMHO but perhaps they implemented it that way, otherwise you could just attach your controller to the onClick() and add some logic to the checkbox (or, this being JS, in a function somewhere) to change the checkbox state. If that's not possible, option 1 and 2 are definitely your best bet.
addition: user interacting with a view
So what happens when a user wants to interact with a view? Frequently a widget will include both a view and the controller. A checkbox has a view (you can see if it's checked or not) and also a controller (you can click it). When you click the checkbox, in principle the following should happen:
checkbox controller receives the event
checkbox controller changes the state for the value this checkbox represents in the model
model updates listeners (including the checkbox)
checkbox updates its look to reflect that that value has changed
The first step, how the controller receives the event is somewhat language dependent, but in OOP languages it's likely a listener object attached to user interface events on this particular widget which either is the controller or notifies the controller of the user interaction.
This is a tough one. If I understand correctly, the problem results because you've exposed a click handler on your model, and the model's click event is caught by the controller. The controller updates the view, which in turn toggles the same event.
From my point of view, I would consider it inappropriate for the Controller to attach itself to the Edge's Click event, because it exposes too much detail about how the Edge is implemented and used. Controller does not care about how the Edge is used or any other implementation details.
In fact, canonical MVC style doesn't require the Controller to hook onto any Model events at all, generally because the Model's state is not mutated by the View or any other Controllers. Its not necessary for the Model to notify the Controller that it's been changed.
To fix your problem, you should define View's interface to have a single method, such as ToggleEdge:
public interface GraphView
{
event Action ToggleEdge;
}
Its tempting to want to create two methods, EdgeClicked and CheckboxClicked, but insisting on two independent methods like that violates the principle of encapsulation. It exposes too many implementation details to your Controller or anyone else who wants to hook onto those events. Remember, the Controller only cares that the View's state changed, it doesn't care how it changed.
When you implement the View interface onto your user interface, you should take care to ensure that the ToggleEdge event is invoked from one location. You can do this by hooking onto the Edge.Clicked event in your View and using it to toggle your checkbox; this makes your checkbox responsible for raising the Toggle vent up to the controller:
public class UI : UserControl, GraphView
{
public event Action ToggleEdge;
void OnToggleEdge(Edge edge)
{
if (ToggleEdge != null)
ToggleEdge(edge);
}
protected void Edge_Clicked(object sender, EventArgs e)
{
CheckBox chkbox = FindCheckBoxThatCorrespondsToEdge((Edge)sender);
chkbox.Checked = !chkbox.Checked;
}
protected void chkEdge_CheckChanged(object sender, EventArgs e)
{
Edge edge = FindEdgeThatCorrespondsToCheckbox((CheckBox)sender);
OnToggleEdge(edge);
}
}
You can make an argument that the View knows too much about its implementation now: its aware that edges and checkboxes are fundamentally connected. Maybe this is another hack, but it can probably be dismissed as "UI logic" need to keep the View's display syncronized.
Related
So some background.
Let's say I have a notice on the page (like a success or error message). This is managed by a collection of Notice models.
When a new notice is added to the collection a region displays it in my page.
But let's now say that I have a view somewhere deep and nested, which when shown should instead display the notice (instead of the other region) until the end of its life.
What method would be clean for managing this?
Adding an attribute to the model to say is_global or something similar
Registering upon view creation that this view will take over responsibility. Then unregistering upon removal.
Some other crazy method
If I am understanding your problem correctly:
You have a queue of notices and depending on the notice they can appear
in different parts of the page
I would say that you should add a property of some sort to the model and, on each add, emit an event related to that property value. This way any view can listen to the event and handle the notice appropriately. This will also keep your notification queue decoupled from the handling view
I'm making a charting application that allows you to create graphs using a drag and drop interface.
I have highcharts and I'm using the highcharts-ng directive.
This directive watches the title, options, and series. And when a person makes a change, I process them and make changes to the options object. Then highcharts-ng redraws the chart.
The problem I'm finding is that I change a few properties in a row such as options.xAxis and options.yAxis, and whenever I do this the application is lagging a bit because it's launching a redraw for every change.
So what would be the angular way to approach this, while still being efficient?
A potential solution I thought of was to add a flag to the highcharts-ng directive, and have it trigger whenever it's changed. And then just change it after I'm done processing the data. Another potential solution is to listen for a certain event inside the highchart-ng directive, and then trigger the redraw whenever that event is received. But these solutions seem/feel a bit hacky to me.
Angular does its own dirty checking and (ideally always, not but really) rewrites the Angular-controlled sections of the DOM whenever their corresponding view models change. I think that this behaviour is so fundamental to Angular that if you don't like it, you'd either better work around it, or use a different databinding framework.
The workaround I'd recommend is basically what you described in your first option: a view model inside the view model. Have a private variable inside the directive's scope which tracks the changes you're interested in, the ones which happen more frequently than you want to redraw. Then when you're ready to redraw (you'll need your own logic for what determines "ready"...time? a particular kind of change? a particular threshold of change?), update the real view model by setting your private variable back to its original field on the real view model.
Code sketch:
// (inside the directive)
var _options = $scope.options;
// ...
// rapidfire updates happen; save them to _options rather than $scope.options
// ...
// now you're ready to redraw:
$scope.options = _options; // angular now knows $scope is dirty and triggers the redraw
I've assembled a modestly sized application and I am in the process of factoring code to reduce the overall number of maintained lines, as well as performance tuning. The use case that has me posting this question is that I have a button embedded in a menu that invokes (or needs to invoke) a method on a controller that displays a form. This currently is implemented using a direct reference to the specific button control, creating a panel, and putting that panel inside of my viewport.
The question at: ExtJS 4.1 Call One Controller From Another begins to address the issue of best-practices near the end of responses, but doesn't really settle on a base-case that can be reproduced or extended to cover more complex implementations (which is the aim of my question.)
Given the two controllers:
A "main menu" controller.
// controller/Menu.js
Ext.define("App.controller.Menu", {
extend: "Ext.app.Controller",
init: function () {
this.control({
"viewport > mainmenu > button": function (ctl, evt) {
}
});
}
});
A user account controller
// controller/User.js
Ext.define("App.controller.User", {
extend: "Ext.app.Controller",
stores: ["User"],
views: ["user.Edit", "user.List"],
init: function () {
}
});
The Question
What would be the (best) way to implement a crosswise connection between the two controllers to properly delegate the responsibility of responding to a click event on the menu button for "Create a New Account?"
One Possible Solution
Using componentquery I can easily narrow down the focus of the button in the main menu view using a tag property such that the User controller is responding directly to the event:
// controller/User.js
"viewport > mainmenu > button [tag=user.create]": function () {}
An unknown alternative
Or I could possible winnow my way through the object graph to find the reference to the User controller from the Menu controller and invoke it that way.
// controller/Menu.js
// switch on tag case "user.create"
App.controller.User.createUserForm()
The real question
The question that results from all of this is what the "most acceptable" solution is here. One could imagine the use of a third, mediating controller, to "route" requests from controls to controllers but I think that goes against what the remainder of the framework is attempting to do. Using a variation of either of these methods currently works however neither feels completely clean and reliable; or ultimately maintainable long-term (as code gets spread out rather quickly.) Additionally the thought had occurred to us to drop into raw events but we run into the same kind of maintainability issues there.
Some short lines:
A thing that I don't understand is that Sencha Touch has routing but no eventbus where ExtJS has a event bus but no routing... (and there are more points where the MVC implementation differ) Whatsoever, because I am using ExtJS most of the time I created a custom routing to fill this gap for me. Maybe sencha will add this in version 5.
The easiest and quickest solution: use the getController() of the Ext.app.Application controller to invoke the responsible controller from your menu controller.
The (imo) best solution: write yourself a router where each controller register it's routes to and use both; routing and eventbus. This gets really handy if your app have shared components that are used by more than one dev team.
Let's say the user might pick one div from the set { A, B, C } and add it to div X, except each div has a different rule that may be dependent the application state and some intrinsic property of each div. Finally upon adding the div, each div might trigger different events.
for example, in one application state, the user:
try add A to X: A is added to X, nothing else happens
try add B to X: B is added to X, some additional event is triggered
try add C to X: not allowed, C is return to some defined original state ( for example if I try to add C to X by drag and drop, C is returned to original position ).
assume all three divs are already rendered and one model, view and controller exist for each one.
Feel free to tell me the pieces of logic you see in this process and where you would put it. But please answer:
should the validation process sit in the controller layer, or somewhere in the model layer
once validated, should one mediator object call the model and view to add new child, or should the model add the child, and then somehow triggers the view to do the same.
should controllers be aware of hierarchal relationships of the dom and model
I am implementing something like this in Ember.js which has certain ideas about things, but please answer even if you don't use it/have never heard of it.
The drag and drop action should be implemented in the view. When the item is dropped it should trigger an action that gets sent to the router. The router (functioning as the app's state machine) should respond to the action appropriately based on the current state. Deciding the appropriate action can be entirely state-based, or take into consideration aspects of the models or controllers.
The action should manipulate the model and/or controllers, and bindings/observers in the view layer should bring the view in sync with the new state of the models/controllers.
Controllers should not be aware of the DOM.
Hope that helps! I cover the division of responsibilities in a presentation available here: http://www.lukemelia.com/blog/archives/2012/08/23/architecting-ember-js-apps/
I have a backbone.js app, whose views have multiple states, which differ substantially from each other ("View","Edit", etc). There are at least 2 different templates for every view. This is OK. My problem is with the JS view managing code.
I rely on an initalize-thin-render-thick approach (which, I think is pretty bad), where the render method is where 80%-90% of the logic occurs. When I want to change the state, I simply call the render method with a specific parameter ("view","edit"). On the basis of that, the view decides what to show and what not, to which events to bind, etc.
I think this is bad, because, on one side it puts bottlenecks on the rendering process, on another, it is not proper state machine, which means that I am not carrying about possible callbacks that might have been bound previously. When I receive the view, I simply clean the view and that's it.
I also observed, that I am not using the delegated event system, provided by backbone, which I think is another minus, because I think, it is very well implemented (BTW, does it make sure to unbind callbacks, when a certain DOM element is removed?)
I think I need some serious refactoring. Please, help with some advice, as to what the best approach for a multi-state Backone view would be.
What I tend to do for these cases is to make a toplevel view that manages a subview for each individual state (index, show, edit, etc.). When a user action is invoked, e.g. "edit this user", "delete this user", "save my changes", the active state view signals the router (directly, or through a hyperlink), and the router will tell the toplevel view to update its state.
Continuing the user editor example, let's say that I have a top level view called UserEditorView. It renders a basic container for the user editor (title bars, etc.) and then, by default, instantiates and renders Users.IndexView inside that container.
Users.IndexView renders the list of users. Next to each user is an edit icon, which is a link to "#users/555/edit". So, when the user clicks it, that event goes to the router, which tells UserEditorView, "hey, I want to edit user #555". And then UserEditorView will remove the IndexView (by calling its .remove() method), instantiate Users.EditView for the appropriate user model, and put the EditView into the container.
When the user is done editing the user, she clicks on "Save", and then EditView saves the model. Now we need to get back to the IndexView. EditView calls window.router.navigate('users', { trigger: true }), so the URL gets updated and the router gets invoked. The router then calls .showIndex() on the UserEditorView, and the UserEditorView does the swap back to IndexView from EditView.
On a simple way to manage unloading of events, I've found this article on zombie views quite useful.
Basically, I don't have a toplevel view, but I render all the views using a view handler that takes care of the views for a given container.
To make your renderer thinner, I would recommend using routes. They are easy to setup, and you can have different views for each route. Or, what I'm used to do is just to have different templates. Using a general Backbone.View overwrite:
Backbone.View = Backbone.View.extend({
initialize: function(attrs) {
attrs = attrs || {}
if(!_.isUndefined(attrs.template)) {
this.template = attrs.template;
}
}
});
I've noticed that I reuse views in two ways:
1. edit views differ only in the underlying model and template, but not the associated logic (clicking the submit validates and saves the model)
2. the same view can be reused in several places with different templates (like a list of users as a ranking or you accounts)
With the above extension, I can pass {template: '/my/current/template/} to the view, and it will be rendered as I want. Together with routes, I finally got a flexible, easy to understand and thin setup.