Consider some pros and cons of the second approach:
0 {{lastUpdated}} instead of {{timerData.lastUpdated}}, which could just as easily be {{timer.lastUpdated}}, which I might argue is more readable (but let's not argue... I'm giving this point a neutral rating so you decide for yourself)
+1 It may be convenient that the controller acts as a sort of API for the markup such that if somehow the structure of the data model changes you can (in theory) update the controller's API mappings without touching the html partial.
-1 However, theory isn't always practice and I usually find myself having to modify markup and controller logic when changes are called for, anyway. So the extra effort of writing the API negates it's advantage.
-1 Furthermore, this approach isn't very DRY.
-1 If you want to bind the data to ng-model your code become even less DRY as you have to re-package the $scope.scalar_values in the controller to make a new REST call.
-0.1 There's a tiny performance hit creating extra watcher(s). Also, if data properties are attached to the model that don't need to be watched in a particular controller they will create additional overhead for the deep watchers.
-1 What if multiple controllers need the same data models? That means that you have multiple API's to update with every model change.
$scope.timerData = Timer.data; is starting to sound mighty tempting right about now... Let's dive a little deeper into that last point... What kind of model changes were we talking about? A model on the back-end (server)? Or a model which is created and lives only in the front-end? In either case, what is essentially the data mapping API belongs in the front-end service layer, (an angular factory or service). (Note that your first example--my preference-- doesn't have such an API in the service layer, which is fine because it's simple enough it doesn't need it.)
In conclusion, everything does not have to be decoupled. And as far as decoupling the markup entirely from the data model, the drawbacks outweigh the advantages.
Controllers, in general shouldn't be littered with $scope = injectable.data.scalar's. Rather, they should be sprinkled with $scope = injectable.data's, promise.then(..)'s, and $scope.complexClickAction = function() {..}'s
As an alternative approach to achieve data-decoupling and thus view-encapsulation, the only place that it really makes sense to decouple the view from the model is with a directive. But even there, don't $watch scalar values in the controller or link functions. That won't save time or make the code any more maintainable nor readable. It won't even make testing easier since robust tests in angular usually test the resulting DOM anyway. Rather, in a directive demand your data API in object form, and favor using just the $watchers created by ng-bind.
Example http://plnkr.co/edit/MVeU1GKRTN4bqA3h9Yio
<body ng-app="ServiceNotification">
<div style="border-style:dotted" ng-controller="TimerCtrl1">
TimerCtrl1<br/>
Bad:<br/>
Last Updated: {{lastUpdated}}<br/>
Last Updated: {{calls}}<br/>
Good:<br/>
Last Updated: {{data.lastUpdated}}<br/>
Last Updated: {{data.calls}}<br/>
</div>
<script src="https://ajax.googleapis.com/ajax/libs/angularjs/1.0.5/angular.js"></script>
<script type="text/javascript">
var app = angular.module("ServiceNotification", []);
function TimerCtrl1($scope, Timer) {
$scope.data = Timer.data;
$scope.lastUpdated = Timer.data.lastUpdated;
$scope.calls = Timer.data.calls;
};
app.factory("Timer", function ($timeout) {
var data = { lastUpdated: new Date(), calls: 0 };
var updateTimer = function () {
data.lastUpdated = new Date();
data.calls += 1;
console.log("updateTimer: " + data.lastUpdated);
$timeout(updateTimer, 500);
};
updateTimer();
return {
data: data
};
});
</script>
</body>
UPDATE: I've finally come back to this question to add that I don't think that either approach is "wrong". Originally I had written that Josh David Miller's answer was incorrect, but in retrospect his points are completely valid, especially his point about separation of concerns.
Separation of concerns aside (but tangentially related), there's another reason for defensive copying that I failed to consider. This question mostly deals with reading data directly from a service. But what if a developer on your team decides that the controller needs to transform the data in some trivial way before the view displays it? (Whether controllers should transform data at all is another discussion.) If she doesn't make a copy of the object first she might unwittingly cause regressions in another view component which consumes the same data.
What this question really highlights are architectural shortcomings of the typical angular application (and really any JavaScript application): tight coupling of concerns, and object mutability. I have recently become enamored with architecting application with React and immutable data structures. Doing so solves the following two problems wonderfully:
Separation of concerns: A component consumes all of it's data via props and has little-to-no reliance on global singletons (such as Angular services), and knows nothing about what happened above it in the view hierarchy.
Mutability: All props are immutable which eliminates the risk of unwitting data mutation.
Angular 2.0 is now on track to borrow heavily from React to achieve the two points above.