As a part of trying to tackle a memory leak in our application, we discovered that for every SkinnableComponent, the skinDestructionPolicy is set to "never" by default.
This means that when using static skin parts, the skin is forever detained in memory.
Furthermore, the override of a partRemoved() in the host component will never be triggered.
Hence, event listeners we add in the partAdded() override are not removed, which effectively causes views and skins to be kept in memory.
When doing a lot of view switches this is just not acceptable.
Here is an example of of how we are working around this now:
public class ViewA extends SkinnableComponent
{
[SkinPart(required = "true")]
public var labelA:Label;
[SkinPart(required = "true")]
public var buttonA:Button;
public function ViewA()
{
super();
mx_internal::skinDestructionPolicy = 'auto';
}
override protected function getCurrentSkinState():String
{
return super.getCurrentSkinState();
}
override protected function partAdded(partName:String, instance:Object):void
{
super.partAdded(partName, instance);
trace("ViewA::partAdded " + partName);
if (instance == buttonA)
{
buttonA.addEventListener(MouseEvent.CLICK, buttonClickedHandler);
}
}
override protected function partRemoved(partName:String, instance:Object):void
{
trace("ViewA::partRemoved " + partName);
if (instance == buttonA)
{
buttonA.removeEventListener(MouseEvent.CLICK, buttonClickedHandler);
}
super.partRemoved(partName, instance);
}
override public function stylesInitialized():void
{
setStyle("skinClass", ViewASkin);
}
}
However, using the mx::internal way to circumvent this behavior seems rather odd to me.
Documentation about this is scarce as well, so any ideas will be very welcome.
Cheers
In my experience the usage of the mx::internal namespace in the Flex SDK usually means something along the lines of: "you can use this functionality, if you know what you're doing, and also we (Adobe, or the Apache community in the future) don't guarantee that this API will never change in future versions of Flex".
So there's no real issue with its usage, unless you're very concerned with backwards compatibility. If you really want to avoid using it, you can always just implement the behavior of skinDestructionPolicy="auto" in your subclass. There's not that much code to write:
override public function initialize():void {
addEventListener(Event.ADDED_TO_STAGE, addedToStageHandler);
addEventListener(Event.REMOVED_FROM_STAGE, removedFromStageHandler);
super.initialize();
}
private function addedToStageHandler(event:Event):void {
if (skin == null) attachSkin();
}
private function removedFromStageHandler(event:Event):void {
detachSkin();
}
Note that in the SkinnableComponent class these two event listeners are attached (or not, depending on the policy) in the commitProperties() method. I moved that to the initialize() method because we don't need to check for changes in the skinDestructionPolicy property anymore.
Also note that this solution might cause an error if one did set mx::internal skinDestructionPolicy to "auto" alongside.
Related
I want to override methods for specific commands on the IDatabase to modify the results.
FT.SEARCH should never return duplicate keys but it can when there is some sort of corruption.
I am using another library that calls FT.SEARCH in IDatabase so this part can only be done by changing IDatabase.
I can autogenerate a wrapper and then use this, but it would need to be maintained and updated every time the IDatabase interface changes. It would be unfortunate to need to update the wrapper class even when I didn't need to override the method that changed in IDatabase. Is there another option that would have similar performance that this wrapper does?
Assume I had a class called DatabaseWrapper all methods were virtual and that was in another package example code would look like:
public sealed class MyDatabaseOverrides: DatabaseWrapper
{
private const string RediSearchSearchCommandName = "FT.SEARCH";
public MyDatabaseOverrides(IDatabase wrapped) : base(wrapped)
{
}
public override Task<RedisResult> ExecuteAsync(string command, params object[] args)
{
var resultTask = base.ExecuteAsync(command, args);
if (string.Equals(RediSearchSearchCommandName, command, StringComparison.OrdinalIgnoreCase))
{
return SkipAndLogDuplicateKeys(resultTask);
}
return resultTask;
}
private async Task<RedisResult> SkipAndLogDuplicateKeys(Task<RedisResult> resultTask)
{
// implementation omitted for brevity
return await resultTask;
}
}
This question already has answers here:
How to make a Java class that implements one interface with two generic types?
(9 answers)
Closed 8 years ago.
I have the following interface, which I want to implement multiple times in my classes:
public interface EventListener<T extends Event>
{
public void onEvent(T event);
}
Now, I want to be able to implement this interface in the following way:
class Foo implements EventListener<LoginEvent>, EventListener<LogoutEvent>
{
#Override
public void onEvent(LoginEvent event)
{
}
#Override
public void onEvent(LogoutEvent event)
{
}
}
However, this gives me the error: Duplicate class com.foo.EventListener on the line:
class Foo implements EventListener<LoginEvent>, EventListener<LogoutEvent>
Is it possible to implement the interface twice with different generics? If not, what's the next closest thing I can do to achieve what I'm trying to do here?
Is it possible to implement the interface twice with different generics
Unfortunately no. The reason you can't implement the same interface twice is because of type erasure. The compiler will handle type parameters, and a runtime EventListener<X> is just a EventListener
If not, what's the next closest thing I can do to achieve what I'm trying to do here?
Type erasure can work in our favor. Once you know that EventListener<X> and EventListener<Y> are just raw EventListener at run-time, it is easier than you think to write an EventListener that can deal with different kinds of Events. Bellow is a solution that passes the IS-A test for EventListener and correctly handles both Login and Logout events by means of simple delegation:
#SuppressWarnings("rawtypes")
public class Foo implements EventListener {
// Map delegation, but could be anything really
private final Map<Class<? extends Event>, EventListener> listeners;
// Concrete Listener for Login - could be anonymous
private class LoginListener implements EventListener<LoginEvent> {
public void onEvent(LoginEvent event) {
System.out.println("Login");
}
}
// Concrete Listener for Logout - could be anonymous
private class LogoutListener implements EventListener<LogoutEvent> {
public void onEvent(LogoutEvent event) {
System.out.println("Logout");
}
}
public Foo() {
#SuppressWarnings("rawtypes")
Map<Class<? extends Event>, EventListener> temp = new HashMap<>();
// LoginEvents will be routed to LoginListener
temp.put(LoginEvent.class, new LoginListener());
// LogoutEvents will be routed to LoginListener
temp.put(LogoutEvent.class, new LogoutListener());
listeners = Collections.unmodifiableMap(temp);
}
#SuppressWarnings("unchecked")
#Override
public void onEvent(Event event) {
// Maps make it easy to delegate, but again, this could be anything
if (listeners.containsKey(event.getClass())) {
listeners.get(event.getClass()).onEvent(event);
} else {
/* Screams if a unsupported event gets passed
* Comment this line if you want to ignore
* unsupported events
*/
throw new IllegalArgumentException("Event not supported");
}
}
public static void main(String[] args) {
Foo foo = new Foo();
System.out.println(foo instanceof EventListener); // true
foo.onEvent(new LoginEvent()); // Login
foo.onEvent(new LogoutEvent()); // Logout
}
}
The suppress warnings are there because we are "abusing" type erasure and delegating to two different event listeners based on the event concrete type. I have chosen to do it using a HashMap and the run-time Event class, but there are a lot of other possible implementations. You could use anonymous inner classes like #user949300 suggested, you could include a getEventType discriminator on the Event class to know what do to with each event and so on.
By using this code for all effects you are creating a single EventListener able to handle two kinds of events. The workaround is 100% self-contained (no need to expose the internal EventListeners).
Finally, there is one last issue that may bother you. At compile time Foo type is actually EventListener. Now, API methods out of your control may be expecting parametrized EventListeners:
public void addLoginListener(EventListener<LoginEvent> event) { // ...
// OR
public void addLogoutListener(EventListener<LogoutEvent> event) { // ...
Again, at run-time both of those methods deal with raw EventListeners. So by having Foo implement a raw interface the compiler will be happy to let you get away with just a type safety warning (which you can disregard with #SuppressWarnings("unchecked")):
eventSource.addLoginListener(foo); // works
While all of this may seem daunting, just repeat to yourself "The compiler is trying to trick me (or save me); there is no spoon <T>. Once you scratch your head for a couple of months trying to make legacy code written before Java 1.5 work with modern code full of type parameters, type erasure becomes second nature to you.
You need to use inner or anonymous classes. For instance:
class Foo {
public EventListener<X> asXListener() {
return new EventListener<X>() {
// code here can refer to Foo
};
}
public EventListener<Y> asYListener() {
return new EventListener<Y>() {
// code here can refer to Foo
};
}
}
This is not possible.
But for that you could create two different classes that implement EventListener interface with two different arguments.
public class Login implements EventListener<LoginEvent> {
public void onEvent(LoginEvent event) {
// TODO Auto-generated method stub
}
}
public class Logout implements EventListener<LogoutEvent> {
public void onEvent(LogoutEvent event) {
// TODO Auto-generated method stub
}
}
I have the following AppDelegate which takes quite some time to load:
Syncfusion.ListView.XForms.iOS.SfListViewRenderer.Init();
new Syncfusion.SfNumericUpDown.XForms.iOS.SfNumericUpDownRenderer();
Syncfusion.SfCarousel.XForms.iOS.SfCarouselRenderer.Init();
Syncfusion.XForms.iOS.Buttons.SfSegmentedControlRenderer.Init();
Syncfusion.XForms.iOS.Buttons.SfCheckBoxRenderer.Init();
new Syncfusion.XForms.iOS.ComboBox.SfComboBoxRenderer();
//Syncfusion.XForms.iOS.TabView.SfTabViewRenderer.Init();
new Syncfusion.SfRotator.XForms.iOS.SfRotatorRenderer();
new Syncfusion.SfRating.XForms.iOS.SfRatingRenderer();
new Syncfusion.SfBusyIndicator.XForms.iOS.SfBusyIndicatorRenderer();
What options should I consider when I know some of these components aren't needed for the main screen, but for subscreens?
I am using PRISM, and it appears that every tab is pre-loaded immediately before allowing display or interaction with the end user. What can I do to delay the pre-rendering that the Prism TabView does prior to showing the interface?
Should I use Lazy<T>? What is the right approach?
Should I move these components to another initialization section?
There are a number of ways you could ultimately achieve this, and it all depends on what your real goals are.
If your goal is to ensure that you get to a Xamarin.Forms Page as fast as possible so that you have some sort of activity indicator, that in essence says to the user, "it's ok I haven't frozen, we're just doing some stuff to get ready for you", then you might try creating a "SpashScreen" page where you do additional loading. The setup might look something like the following:
public partial class AppDelegate : FormsApplicationDelegate
{
public override bool FinishedLaunching(UIApplication app, NSDictionary options)
{
global::Xamarin.Forms.Forms.Init();
LoadApplication(new App(new iOSInitializer()));
return base.FinishedLaunching(app, options);
}
}
}
public class iOSInitializer : IPlatformInitializer, IPlatformFinalizer
{
public void RegisterTypes(IContainerRegistry containerRegistry)
{
containerRegistry.RegisterInstance<IPlatformFinalizer>(this);
}
public void Finalize()
{
new Syncfusion.SfNumericUpDown.XForms.iOS.SfNumericUpDownRenderer();
Syncfusion.SfCarousel.XForms.iOS.SfCarouselRenderer.Init();
Syncfusion.XForms.iOS.Buttons.SfSegmentedControlRenderer.Init();
Syncfusion.XForms.iOS.Buttons.SfCheckBoxRenderer.Init();
}
}
public class App : PrismApplication
{
protected override async void OnInitialized()
{
await NavigationService.NavigateAsync("SplashScreen");
}
}
public class SplashScreenViewModel : INavigationAware
{
private IPlatformFinalizer _platformFinalizer { get; }
private INavigationService _navigationService { get; }
public SplashScreenViewModel(INavigationService navigationService, IPlatformFinalizer platformFinalizer)
{
_navigationService = navigationService;
_platformFinalizer = platformFinalizer;
}
public async void OnNavigatedTo(INavigationParameters navigationParameters)
{
_platformFinalizer.Finalize();
await _navigationService.NavigateAsync("/MainPage");
}
}
If you're working with Modules you could take a similar approach though any Modules that would initialize at Startup would still be making that call to Init the renderers before you've set a Page to navigate to. That said, working with Modules does give you a number of benefits here as you only ever would have to initialize things that the app actually requires at that point.
All of that said I'd be surprised if you see much in the way of gain as these Init calls are typically empty methods only designed to prevent the Linker from linking them out... if you aren't linking or have a linker file you could simply instruct the Linker to leave your Syncfusion and other libraries alone.
This question is posed from an ActionScript context but could be from a Java one equally.
The code I am using as a sample comes from adobe.com/devnet/flex/articles/flex4_skinning.html. In the code extract below the NoteCard class has an enabled and a disabled state which it inherits from the SkinnableComponent class. My question is; why in the enabled setter do we call super.enabled = value; and not this.enabled = value;. We have created our NoteCard object instance from the constructor and should we not then be able to set the value of the enabled member using the "this" keyword. If you do swap super for this no errors are shown by the compiler but the code fails to work.
package
{
import spark.components.supportClasses.SkinnableComponent;
public class NoteCard extends SkinnableComponent
{
public function NoteCard()
{
super();
}
override public function set enabled(value:Boolean) : void
{
if (enabled != value)
invalidateSkinState();
super.enabled = value;
}
override protected function getCurrentSkinState() : String
{
if (!enabled)
return "disabled";
return "normal"
}
}
}
If we'll use:
override public function set enabled(value:Boolean) : void
{
if (enabled != value)
invalidateSkinState();
enabled = value;
}
We'll run into infinite loop. This line:
enabled = value;
will call the same setter again and again.
In this special case you override a setter for a class. You can implement your own additional code to handle the newly set value, but the code from your superclass should also be called, because you may not know, what the base class setter will do (may be set a private variable with a value). You have to call super.setterName = value to assure this. If you would call with this you would call your implemented setter in an infinite loop. You may omit the super call, if you are sure, this isn't necessary.
Being new to RhinoMocks and Unit Testing, I have come accross an issue that I cannot seem to find a resolution to (no matter how much documentation I read).
The issue is this: I have created an Interface that exposes 5 Events (to be used for a view in ASP.NET and the MVP Supervisory Controller pattern..... I know, I should be using MVC, but that's a whole other issue). Anyway, I want to test that when a certain event fires on the view, we'll call it "IsLoaded", that a method inside of my Presenter is called and, using Dependency Injection, a value is returned from the Dependency and set to the view. Here is where the problem starts: when I use Expect.Call(Dependency.GetInfo()).Return(SomeList), the Call never executes (without the mock.ReplayAll() method being invoked). Well, when I invoke the ReplayAll method, I get ExpectationExceptions because of the Subscription by the Presenter object to the other Events exposed by the View Interface.
So, for me to test that IView.IsLoaded has fired, I want to verify that IView.ListOfSomething has been updated to match the list I passed in via the Expect.Call(). However, when I set the expectation, the other Event subscriptions (which occur straight out of the constructor for the Presenter) fail the #0 Expectations of the test. What I get is, view.Save += this.SaveNewList tosses up a RhinoMocks ExpectationViolationException.
My million dollar question is this: Is it necessary I set expectations for ALL of my events (via [Setup]), or is there something that I'm missing/not understanding about how Unit Testing or RhinoMocks works?
Please bear in mind I am extremely new to Unit Testing, and therefore RhinoMocks. If it appears I don't know what I'm talking about, please feel free to point that out.
I'm working on a project where we used MVP and rhino mocks as well. What we did was simply expect all event subscriptions in every test.
private void SetupDefaultExpectations()
{
_mockView.Initializing += null; LastCall.IgnoreArguments();
_mockView.SavingChanges += null; LastCall.IgnoreArguments();
}
Then we built a extension method on IMockedObject (from RhinoMocks) to trigger events in the unit tests and un-wrap exceptions so that they can be expected in the standard NUnit way.
static class IMockedObjectExtension
{
public static void RaiseEvent(this IMockedObject mockView, string eventName, EventArgs args)
{
EventRaiser eventraiser = new EventRaiser(mockView, eventName);
try
{
eventraiser.Raise(mockView, args);
}
catch (TargetInvocationException ex)
{
throw ex.InnerException;
}
}
public static void RaiseEvent(this IMockedObject mockView, string eventName)
{
RaiseEvent(mockView, eventName, EventArgs.Empty);
}
}
This could then be used from the unit test like this
using(_mocks.Record())
{
Expect.Call(dependency.GetInfo()).Return(someList);
}
using(_mocks.Playback())
{
Presenter presenter = new Presenter(_mockView, dependency);
(_mockView as IMockedObject).RaiseEvent("SavingChanges");
}
To eliminate duplication between presenter tests we have refactored this to a BasePresenterTest base class which sets up this basic structure for all presenter tests and exposes helper methods to the sub class.
public abstract class BasePresenterTest<VIEW> where VIEW : IBaseView
{
protected MockRepository _mocks;
protected VIEW View { get; private set; }
protected abstract void SetUp();
protected abstract void TearDown();
protected abstract void SetupDefaultExpectations();
[SetUp]
public virtual void BaseSetUp()
{
_mocks = new MockRepository();
View = _mocks.CreateMock<VIEW>();
SetUp();
}
[TearDown]
public virtual void BaseTearDown()
{
TearDown();
View = null;
_mocks = null;
}
protected virtual void BaseSetupDefaultExpectations()
{
//Setup default expectations that are general for all views
SetupDefaultExpectations();
}
protected virtual IDisposable Record()
{
IDisposable mocksRecordState = _mocks.Record();
BaseSetupDefaultExpectations();
return mocksRecordState;
}
protected virtual IDisposable Playback()
{
return _mocks.Playback();
}
protected void RaiseEventOnView(string eventName)
{
(View as IMockedObject).RaiseEvent(eventName);
}
}
This eliminates alot of code from the tests in our project.
We still use a old version of RhinoMocks but I will try to update this once we move to a later version.