How do I synchronize this properly? - android

public void consumeResponse(OmwListResponse<T> response) {
synchronized (response.getResultList()) { // XXX this isn't synchronized safely
for (T t : response.getResultList()) {
if (!cacheList.contains(t)) {
cacheList.add(t);
}
}
}
}
The the situation is I don't want anyone to chance response.getResultList() or cacheList until this method is done. How do I properly do this?

Create a lock object:
private static final void LOCK = new Object();
and synchronize on that.

Synchronizing cacheList is easy. Just wrap any code your code that uses it in:
synchronized(cacheList) {
// Make changes to cacheList here
}
If cacheList is a public member and you're afraid external classes will change it, make it a private member and synchronize the getter and setter. This is the only way since you have no control over what other classes do and it is your responsibility to synchronize your members.
As for response, that is trickier because I don't what an OmwListResponse is. Do you own that class? If so, use the same method as above. If not, you may be out of luck.

Related

Using PixelCopy to take a screenshot [duplicate]

What is the use of anonymous classes in Java? Can we say that usage of anonymous class is one of the advantages of Java?
By an "anonymous class", I take it you mean anonymous inner class.
An anonymous inner class can come useful when making an instance of an object with certain "extras" such as overriding methods, without having to actually subclass a class.
I tend to use it as a shortcut for attaching an event listener:
button.addActionListener(new ActionListener() {
#Override
public void actionPerformed(ActionEvent e) {
// do something
}
});
Using this method makes coding a little bit quicker, as I don't need to make an extra class that implements ActionListener -- I can just instantiate an anonymous inner class without actually making a separate class.
I only use this technique for "quick and dirty" tasks where making an entire class feels unnecessary. Having multiple anonymous inner classes that do exactly the same thing should be refactored to an actual class, be it an inner class or a separate class.
Anonymous inner classes are effectively closures, so they can be used to emulate lambda expressions or "delegates". For example, take this interface:
public interface F<A, B> {
B f(A a);
}
You can use this anonymously to create a first-class function in Java. Let's say you have the following method that returns the first number larger than i in the given list, or i if no number is larger:
public static int larger(final List<Integer> ns, final int i) {
for (Integer n : ns)
if (n > i)
return n;
return i;
}
And then you have another method that returns the first number smaller than i in the given list, or i if no number is smaller:
public static int smaller(final List<Integer> ns, final int i) {
for (Integer n : ns)
if (n < i)
return n;
return i;
}
These methods are almost identical. Using the first-class function type F, we can rewrite these into one method as follows:
public static <T> T firstMatch(final List<T> ts, final F<T, Boolean> f, T z) {
for (T t : ts)
if (f.f(t))
return t;
return z;
}
You can use an anonymous class to use the firstMatch method:
F<Integer, Boolean> greaterThanTen = new F<Integer, Boolean> {
Boolean f(final Integer n) {
return n > 10;
}
};
int moreThanMyFingersCanCount = firstMatch(xs, greaterThanTen, x);
This is a really contrived example, but its easy to see that being able to pass functions around as if they were values is a pretty useful feature. See "Can Your Programming Language Do This" by Joel himself.
A nice library for programming Java in this style: Functional Java.
Anonymous inner class is used in following scenario:
1.) For Overriding(subclassing), when class definition is not usable except current case:
class A{
public void methodA() {
System.out.println("methodA");
}
}
class B{
A a = new A() {
public void methodA() {
System.out.println("anonymous methodA");
}
};
}
2.) For implementing an interface, when implementation of interface is required only for current case:
interface InterfaceA{
public void methodA();
}
class B{
InterfaceA a = new InterfaceA() {
public void methodA() {
System.out.println("anonymous methodA implementer");
}
};
}
3.) Argument Defined Anonymous inner class:
interface Foo {
void methodFoo();
}
class B{
void do(Foo f) { }
}
class A{
void methodA() {
B b = new B();
b.do(new Foo() {
public void methodFoo() {
System.out.println("methodFoo");
}
});
}
}
I use them sometimes as a syntax hack for Map instantiation:
Map map = new HashMap() {{
put("key", "value");
}};
vs
Map map = new HashMap();
map.put("key", "value");
It saves some redundancy when doing a lot of put statements. However, I have also run into problems doing this when the outer class needs to be serialized via remoting.
They're commonly used as a verbose form of callback.
I suppose you could say they're an advantage compared to not having them, and having to create a named class every time, but similar concepts are implemented much better in other languages (as closures or blocks)
Here's a swing example
myButton.addActionListener(new ActionListener(){
public void actionPerformed(ActionEvent e) {
// do stuff here...
}
});
Although it's still messily verbose, it's a lot better than forcing you to define a named class for every throw away listener like this (although depending on the situation and reuse, that may still be the better approach)
You use it in situations where you need to create a class for a specific purpose inside another function, e.g., as a listener, as a runnable (to spawn a thread), etc.
The idea is that you call them from inside the code of a function so you never refer to them elsewhere, so you don't need to name them. The compiler just enumerates them.
They are essentially syntactic sugar, and should generally be moved elsewhere as they grow bigger.
I'm not sure if it is one of the advantages of Java, though if you do use them (and we all frequently use them, unfortunately), then you could argue that they are one.
GuideLines for Anonymous Class.
Anonymous class is declared and initialized simultaneously.
Anonymous class must extend or implement to one and only one class or interface resp.
As anonymouse class has no name, it can be used only once.
eg:
button.addActionListener(new ActionListener(){
public void actionPerformed(ActionEvent arg0) {
// TODO Auto-generated method stub
}
});
Yes, anonymous inner classes is definitely one of the advantages of Java.
With an anonymous inner class you have access to final and member variables of the surrounding class, and that comes in handy in listeners etc.
But a major advantage is that the inner class code, which is (at least should be) tightly coupled to the surrounding class/method/block, has a specific context (the surrounding class, method, and block).
new Thread() {
public void run() {
try {
Thread.sleep(300);
} catch (InterruptedException e) {
System.out.println("Exception message: " + e.getMessage());
System.out.println("Exception cause: " + e.getCause());
}
}
}.start();
This is also one of the example for anonymous inner type using thread
An inner class is associated with an instance of the outer class and there are two special kinds: Local class and Anonymous class. An anonymous class enables us to declare and instantiate a class at same time, hence makes the code concise. We use them when we need a local class only once as they don't have a name.
Consider the example from doc where we have a Person class:
public class Person {
public enum Sex {
MALE, FEMALE
}
String name;
LocalDate birthday;
Sex gender;
String emailAddress;
public int getAge() {
// ...
}
public void printPerson() {
// ...
}
}
and we have a method to print members that match search criteria as:
public static void printPersons(
List<Person> roster, CheckPerson tester) {
for (Person p : roster) {
if (tester.test(p)) {
p.printPerson();
}
}
}
where CheckPerson is an interface like:
interface CheckPerson {
boolean test(Person p);
}
Now we can make use of anonymous class which implements this interface to specify search criteria as:
printPersons(
roster,
new CheckPerson() {
public boolean test(Person p) {
return p.getGender() == Person.Sex.MALE
&& p.getAge() >= 18
&& p.getAge() <= 25;
}
}
);
Here the interface is very simple and the syntax of anonymous class seems unwieldy and unclear.
Java 8 has introduced a term Functional Interface which is an interface with only one abstract method, hence we can say CheckPerson is a functional interface. We can make use of Lambda Expression which allows us to pass the function as method argument as:
printPersons(
roster,
(Person p) -> p.getGender() == Person.Sex.MALE
&& p.getAge() >= 18
&& p.getAge() <= 25
);
We can use a standard functional interface Predicate in place of the interface CheckPerson, which will further reduce the amount of code required.
i use anonymous objects for calling new Threads..
new Thread(new Runnable() {
public void run() {
// you code
}
}).start();
Anonymous inner class can be beneficial while giving different implementations for different objects. But should be used very sparingly as it creates problem for program readability.
One of the major usage of anonymous classes in class-finalization which called finalizer guardian. In Java world using the finalize methods should be avoided until you really need them. You have to remember, when you override the finalize method for sub-classes, you should always invoke super.finalize() as well, because the finalize method of super class won't invoke automatically and you can have trouble with memory leaks.
so considering the fact mentioned above, you can just use the anonymous classes like:
public class HeavyClass{
private final Object finalizerGuardian = new Object() {
#Override
protected void finalize() throws Throwable{
//Finalize outer HeavyClass object
}
};
}
Using this technique you relieved yourself and your other developers to call super.finalize() on each sub-class of the HeavyClass which needs finalize method.
You can use anonymous class this way
TreeSet treeSetObj = new TreeSet(new Comparator()
{
public int compare(String i1,String i2)
{
return i2.compareTo(i1);
}
});
Seems nobody mentioned here but you can also use anonymous class to hold generic type argument (which normally lost due to type erasure):
public abstract class TypeHolder<T> {
private final Type type;
public TypeReference() {
// you may do do additional sanity checks here
final Type superClass = getClass().getGenericSuperclass();
this.type = ((ParameterizedType) superClass).getActualTypeArguments()[0];
}
public final Type getType() {
return this.type;
}
}
If you'll instantiate this class in anonymous way
TypeHolder<List<String>, Map<Ineger, Long>> holder =
new TypeHolder<List<String>, Map<Ineger, Long>>() {};
then such holder instance will contain non-erasured definition of passed type.
Usage
This is very handy for building validators/deserializators. Also you can instantiate generic type with reflection (so if you ever wanted to do new T() in parametrized type - you are welcome!).
Drawbacks/Limitations
You should pass generic parameter explicitly. Failing to do so will lead to type parameter loss
Each instantiation will cost you additional class to be generated by compiler which leads to classpath pollution/jar bloating
An Anonymous Inner Class is used to create an object that will never be referenced again. It has no name and is declared and created in the same statement.
This is used where you would normally use an object's variable. You replace the variable with the new keyword, a call to a constructor and the class definition inside { and }.
When writing a Threaded Program in Java, it would usually look like this
ThreadClass task = new ThreadClass();
Thread runner = new Thread(task);
runner.start();
The ThreadClass used here would be user defined. This class will implement the Runnable interface which is required for creating threads. In the ThreadClass the run() method (only method in Runnable) needs to be implemented as well.
It is clear that getting rid of ThreadClass would be more efficient and that's exactly why Anonymous Inner Classes exist.
Look at the following code
Thread runner = new Thread(new Runnable() {
public void run() {
//Thread does it's work here
}
});
runner.start();
This code replaces the reference made to task in the top most example. Rather than having a separate class, the Anonymous Inner Class inside the Thread() constructor returns an unnamed object that implements the Runnable interface and overrides the run() method. The method run() would include statements inside that do the work required by the thread.
Answering the question on whether Anonymous Inner Classes is one of the advantages of Java, I would have to say that I'm not quite sure as I am not familiar with many programming languages at the moment. But what I can say is it is definitely a quicker and easier method of coding.
References: Sams Teach Yourself Java in 21 Days Seventh Edition
The best way to optimize code. also, We can use for an overriding method of a class or interface.
import java.util.Scanner;
abstract class AnonymousInner {
abstract void sum();
}
class AnonymousInnerMain {
public static void main(String []k){
Scanner sn = new Scanner(System.in);
System.out.println("Enter two vlaues");
int a= Integer.parseInt(sn.nextLine());
int b= Integer.parseInt(sn.nextLine());
AnonymousInner ac = new AnonymousInner(){
void sum(){
int c= a+b;
System.out.println("Sum of two number is: "+c);
}
};
ac.sum();
}
}
One more advantage:
As you know that Java doesn't support multiple inheritance, so if you use "Thread" kinda class as anonymous class then the class still has one space left for any other class to extend.

Weak reference being nullified

I'm using a weak reference inside a static Handler to avoid memory leaks, however, sometimes this reference is being nullified, I cannot understand why.
The static handler is defined inside a repository class that has a method to perform an operation in the background, receives a callback to notify the caller when it's done:
public class MyRepository {
public void performOperation(ContentResolver cr, RepositoryCallback callback) {
MyHandler handler = new MyHandler(cr, callback);
handler.startQuery(...)
}
interface RepositoryCallback {
void onSuccess(MyModel model);
}
// Handler class code here
}
The code of the handler is the following:
private static class MyHandler extends AsyncQueryHandler {
private final WeakReference<RepositoryCallback> weakCallback;
public MyHandler(ContentResolver cr, RepositoryCallback callback) {
super(cr);
this.weakCallback = new WeakReference<>(callback);
}
#Override
protected void onQueryComplete(int token, Object cookie, Cursor cursor) {
RepositoryCallback callback = this.weakCallback.get();
if (callback != null) { // --> Here sometimes it is null
// Do some stuff with the cursor to create MyModel
callback.onSuccess(model);
}
}
}
For some reason, this.weakCallback.get() sometimes is null, and I'm trying to understand why.
The activity code looks like this:
public class MyActivity extends AppCompatActivity {
public void loadModel() {
showLoadingView();
myRepository.performOperation(context.getContentResolver(), new RepositoryCallback() {
#Override
public void onSuccess(MyModel model) {
hideLoadingView();
// Do something with model
}
});
}
}
As you can see I'm creating an anonymous class for the callback, but nobody is holding a reference to it.
Is this the cause of the weak reference being nullified?
Thanks.
That's the "classical" bug associated with weak references.
If the Observable holds the only reference to the Observer, and this reference is weak, then it can be cleared and Observer be garbage collected.
Since you're using anonymous class, Observable will hold the only reference to it, therefore it will be cleared.
As a side note - in my entire experience of Android development, whenever I saw devs using weak references, it always was a code smell. Usually it indicates that either devs don't understand how weak references work, or they don't trust their own code.
A good rule of thumb is that you should never use weak references.
EDIT:
I think that Handler is an anti-pattern in general. You can read more about this in this Reddit thread. There is also a thread there in which I helped one dev to see how he can get rid of HandlerThread in his codebase.
On the other hand, Jake Wharton disagreed with my statements.
Take what you'd like from there, but, in general, I would say that having a static Handler is anti-pattern for sure.
If you are worried about AndroidStudion warnings, then just remember that Google are responsible for AsyncTask and Loaders. This warning is not just useless, but actually bad. They should've made it you should not use static Hadlers.
If all you need is to offload work to BG thread and then get a callback on UI thread then you would be much better off with something like RxJava. Or even the evil AsyncTask.
I guess you're using AsyncQueryHandler in order to access ContentProvider. This is too a very controversial approach. If you don't need to share data with other apps, you might be better off by using some ORM that handles the multithreading for you.

Test void method using mokito

How do I test below method using mockito
public void showArg(String ss) {
if(ss == null) {
throw new NullPointerException();
}else if(ss.equals("")) {
throw new IllegalArgumentException();
}
// Log.d("",""+ss);
if(ss.equals("xyz")) {
this.show();
}else {
getResult(0);
}
}
In this example, there is nothing to be mocked. I just want to test the that is appropriate methods are called based on i/p.
If you want to verify that this method was called (assuming it was public), I suggest using a spy...
MyClass spy = Mockito.spy( myActualObject );
spy.showArg("xyz");
Mockito.verify(spy).show();
Spying (instead of mocking) means to take an actual object and "spy" on it, by wrapping it in another instance. This way you can call actual methods, but also check what was called and even modify what some methods will do, similar to mocking (the difference is, that a mock does not have an underlying "real" object, while a spy has).
As already mentioned you should use a spy to test such code. Additionaly looking at your code you should also test whether appropiate exceptions are thrown.
Code testing border cases can be looking like this:
#Test(expected = NullPointerException.class)
public void shouldThrowNullPointerExceptionWhenNullStringProvided() {
showArg(null);
}
#Test(expected = IllegalArgumentException.class)
public void shouldThrowIllegarArgumentExceptionWhenEmptyStringProvided() {
showArg("");
}

Android threads locking using synchronized

I'm trying to use locks in Android to detect when the activity is active or not, the flow of my app is below,
I'm using threads running by service and I have one activity.
The thread will check if the Activity is open or not,
1- If it is open/active then it will NOT update the DB.
2- If the activity/active isn't open then it will update the DB.
Is the below code correct,
On the activity side I'm doing this,
public void onCreate(Bundle savedInstanceState) {
GlobalVars.setActivityActive()
}
On the thread side I'm doing this,
if (GlobalVars.isActivityActive())
do nothing
else
update DB
And the global vars class is this
public class GlobalVars extends Application {
private boolean ActivityActive;
public boolean isActivityActive() {
synchronized (this) {
return ActivityActive;
}
}
public void setActivityActive(boolean ActivityActive) {
synchronized (this) {
this.ActivityActive = ActivityActive;
}
}
}
My question is this, is using synchronized in the get correct? or should I just do this,
public boolean isActivityActive() {
return ActivityActive;
}
Update
This is how I ended up using it based on the answers/comments
private volatile boolean activityActive;// this would be enough to be thread safe
Yeah, that will work.
It's crazy, though. If you are gonna synchronize on "this" you could make it simpler by just synchronizing the methods. ... so what is that ReentrantLock for?
... and you could make the whole thing even simpler by just using an AtomicBoolean.

Best Practices for Organizing Android Code

I have been coding an Android app that has a lot of code dedicated to it. As you can imagine, there's lots of case-driven code in there. Because most of Android callback functionality is based on integers and ItemIDs and requestCodes, there is a lot of functionality built into switch statements or if-then-else constructs.
What are the best practices for organizing/refactoring this code in a better way? What have you found that works to reduce the amount of code and clarifies it at the same time? Is a huge amount of small classes going to hurt Android performance?
Thanks in advance.
A large number of classes is not going to affect the performance of the application. Some good practices in Android, however, include placing values like integers, item IDs, and request codes into a Resources xml file.
You will also see a lot of Callback classes as inner interfaces of the Object they relate to:
public class MyObject
{
private Callback callback;
private Object returnObject;
public void setCallback(Callback callback)
{
this.callback = callback;
}
public void doSomething()
{
//do something - could be an anync task or other that assigns returnObject
callback.invoke(returnObject);
}
public interface Callback
{
public void invoke(Object obj);
}
}
Then you can use this as follows:
MyObject obj = new MyObject();
obj.setCallback(new MyObject.Callback() {
#Override
public void invoke(Object obj) {
Log.i("Callback", obj.toString());
}
});
obj.doSomething();

Categories

Resources