Angular Directive: when to use link, and when to use scope? - angularjs

Question

When building angular directives, I've found there is more than one way to pass a value from the parent scope to the child scope. Some ways I'm aware of are:
Don't use an isolate scope at all, and the child will simply have
access to the parent scope (you can mount a pretty good argument that this is bad).
Use the attributes parameter of a link function.
Use an isolate scope and bind to the attribute (e.g. param: '=')
The codepen here: https://codepen.io/ariscol/pen/WEKzMe shows two similar directives, one done with link and one done with 2-way binding in an isolate scope. Furthermore, it shows how they differ as far as 1-time binding compared to 2-way binding. For reference, here are the two directives:
app.directive("contactWidgetWithScope", function() {
return {
restrict: 'E',
template: '<div ng-bind="contact.name"></div>'
+ '<div ng-bind="contact.title"></div>'
+ '<div ng-bind="contact.phone"></div>',
scope: {
contact: '='
}
};
});
app.directive("contactWidgetWithLink", function() {
return {
restrict: 'E',
template: '<div ng-bind="name"></div>'
+ '<div ng-bind="title"></div>'
+ '<div ng-bind="phone"></div>',
scope: {},
link: function(scope, elem, attrs) {
scope.name = attrs.contactname;
scope.title = attrs.contacttitle;
scope.phone = attrs.contactphone;
}
};
});
Now, if I were trying to decide which way was "better", I might consider how I was going to use this directive. If I was going to have a thousand contacts, and I wanted to use this directive to list all one thousand contacts on a page, in an ng-repeat, for example, I imagine that I would have significantly better performance with link, as it won't add any watchers. On the other hand, if I wanted this directive to be incorporated into a page header, and I wanted the contact details to be updated as you clicked on any given contact in a list, I would want 2-way binding, so that any change to some "selectedContact" property in a parent scope would be automatically reflected in this directive. Are those the proper considerations? Are there others?
To add to my confusion, it is simple to add an observer to a linked attribute and achieve a 1-way binding such that a change in the value of the attribute will be reflected in the child. Would doing this have more or less of a performance impact? Conversely, I imagine you could do a 1-time binding on the value of the scope version and thereby eliminate the performance impact, e.g.: <contact-widget-with-scope contact="::vm.contact">. That should work, right? Seems like that option gives you a lot of flexibility, because it means the person who invokes the directive can decide if they want to pay the performance price to get the benefit of 2-way binding or not. Are these considerations accurate? Are there other things I ought to consider when deciding how to make values available to my directives?