Related
I read this question and thought that would easily be solved (not that it isn't solvable without) if one could write:
#Override
public String toString() {
return super.super.toString();
}
I'm not sure if it is useful in many cases, but I wonder why it isn't and if something like this exists in other languages.
What do you guys think?
EDIT:
To clarify: yes I know, that's impossible in Java and I don't really miss it. This is nothing I expected to work and was surprised getting a compiler error. I just had the idea and like to discuss it.
It violates encapsulation. You shouldn't be able to bypass the parent class's behaviour. It makes sense to sometimes be able to bypass your own class's behaviour (particularly from within the same method) but not your parent's. For example, suppose we have a base "collection of items", a subclass representing "a collection of red items" and a subclass of that representing "a collection of big red items". It makes sense to have:
public class Items
{
public void add(Item item) { ... }
}
public class RedItems extends Items
{
#Override
public void add(Item item)
{
if (!item.isRed())
{
throw new NotRedItemException();
}
super.add(item);
}
}
public class BigRedItems extends RedItems
{
#Override
public void add(Item item)
{
if (!item.isBig())
{
throw new NotBigItemException();
}
super.add(item);
}
}
That's fine - RedItems can always be confident that the items it contains are all red. Now suppose we were able to call super.super.add():
public class NaughtyItems extends RedItems
{
#Override
public void add(Item item)
{
// I don't care if it's red or not. Take that, RedItems!
super.super.add(item);
}
}
Now we could add whatever we like, and the invariant in RedItems is broken.
Does that make sense?
I think Jon Skeet has the correct answer. I'd just like to add that you can access shadowed variables from superclasses of superclasses by casting this:
interface I { int x = 0; }
class T1 implements I { int x = 1; }
class T2 extends T1 { int x = 2; }
class T3 extends T2 {
int x = 3;
void test() {
System.out.println("x=\t\t" + x);
System.out.println("super.x=\t\t" + super.x);
System.out.println("((T2)this).x=\t" + ((T2)this).x);
System.out.println("((T1)this).x=\t" + ((T1)this).x);
System.out.println("((I)this).x=\t" + ((I)this).x);
}
}
class Test {
public static void main(String[] args) {
new T3().test();
}
}
which produces the output:
x= 3
super.x= 2
((T2)this).x= 2
((T1)this).x= 1
((I)this).x= 0
(example from the JLS)
However, this doesn't work for method calls because method calls are determined based on the runtime type of the object.
I think the following code allow to use super.super...super.method() in most case.
(even if it's uggly to do that)
In short
create temporary instance of ancestor type
copy values of fields from original object to temporary one
invoke target method on temporary object
copy modified values back to original object
Usage :
public class A {
public void doThat() { ... }
}
public class B extends A {
public void doThat() { /* don't call super.doThat() */ }
}
public class C extends B {
public void doThat() {
Magic.exec(A.class, this, "doThat");
}
}
public class Magic {
public static <Type, ChieldType extends Type> void exec(Class<Type> oneSuperType, ChieldType instance,
String methodOfParentToExec) {
try {
Type type = oneSuperType.newInstance();
shareVars(oneSuperType, instance, type);
oneSuperType.getMethod(methodOfParentToExec).invoke(type);
shareVars(oneSuperType, type, instance);
} catch (Exception e) {
throw new RuntimeException(e);
}
}
private static <Type, SourceType extends Type, TargetType extends Type> void shareVars(Class<Type> clazz,
SourceType source, TargetType target) throws IllegalArgumentException, IllegalAccessException {
Class<?> loop = clazz;
do {
for (Field f : loop.getDeclaredFields()) {
if (!f.isAccessible()) {
f.setAccessible(true);
}
f.set(target, f.get(source));
}
loop = loop.getSuperclass();
} while (loop != Object.class);
}
}
I don't have enough reputation to comment so I will add this to the other answers.
Jon Skeet answers excellently, with a beautiful example. Matt B has a point: not all superclasses have supers. Your code would break if you called a super of a super that had no super.
Object oriented programming (which Java is) is all about objects, not functions. If you want task oriented programming, choose C++ or something else. If your object doesn't fit in it's super class, then you need to add it to the "grandparent class", create a new class, or find another super it does fit into.
Personally, I have found this limitation to be one of Java's greatest strengths. Code is somewhat rigid compared to other languages I've used, but I always know what to expect. This helps with the "simple and familiar" goal of Java. In my mind, calling super.super is not simple or familiar. Perhaps the developers felt the same?
There's some good reasons to do this. You might have a subclass which has a method which is implemented incorrectly, but the parent method is implemented correctly. Because it belongs to a third party library, you might be unable/unwilling to change the source. In this case, you want to create a subclass but override one method to call the super.super method.
As shown by some other posters, it is possible to do this through reflection, but it should be possible to do something like
(SuperSuperClass this).theMethod();
I'm dealing with this problem right now - the quick fix is to copy and paste the superclass method into the subsubclass method :)
In addition to the very good points that others have made, I think there's another reason: what if the superclass does not have a superclass?
Since every class naturally extends (at least) Object, super.whatever() will always refer to a method in the superclass. But what if your class only extends Object - what would super.super refer to then? How should that behavior be handled - a compiler error, a NullPointer, etc?
I think the primary reason why this is not allowed is that it violates encapsulation, but this might be a small reason too.
I think if you overwrite a method and want to all the super-class version of it (like, say for equals), then you virtually always want to call the direct superclass version first, which one will call its superclass version in turn if it wants.
I think it only makes rarely sense (if at all. i can't think of a case where it does) to call some arbitrary superclass' version of a method. I don't know if that is possible at all in Java. It can be done in C++:
this->ReallyTheBase::foo();
At a guess, because it's not used that often. The only reason I could see using it is if your direct parent has overridden some functionality and you're trying to restore it back to the original.
Which seems to me to be against OO principles, since the class's direct parent should be more closely related to your class than the grandparent is.
Calling of super.super.method() make sense when you can't change code of base class. This often happens when you are extending an existing library.
Ask yourself first, why are you extending that class? If answer is "because I can't change it" then you can create exact package and class in your application, and rewrite naughty method or create delegate:
package com.company.application;
public class OneYouWantExtend extends OneThatContainsDesiredMethod {
// one way is to rewrite method() to call super.method() only or
// to doStuff() and then call super.method()
public void method() {
if (isDoStuff()) {
// do stuff
}
super.method();
}
protected abstract boolean isDoStuff();
// second way is to define methodDelegate() that will call hidden super.method()
public void methodDelegate() {
super.method();
}
...
}
public class OneThatContainsDesiredMethod {
public void method() {...}
...
}
For instance, you can create org.springframework.test.context.junit4.SpringJUnit4ClassRunner class in your application so this class should be loaded before the real one from jar. Then rewrite methods or constructors.
Attention: This is absolute hack, and it is highly NOT recommended to use but it's WORKING! Using of this approach is dangerous because of possible issues with class loaders. Also this may cause issues each time you will update library that contains overwritten class.
#Jon Skeet Nice explanation.
IMO if some one wants to call super.super method then one must be want to ignore the behavior of immediate parent, but want to access the grand parent behavior.
This can be achieved through instance Of. As below code
public class A {
protected void printClass() {
System.out.println("In A Class");
}
}
public class B extends A {
#Override
protected void printClass() {
if (!(this instanceof C)) {
System.out.println("In B Class");
}
super.printClass();
}
}
public class C extends B {
#Override
protected void printClass() {
System.out.println("In C Class");
super.printClass();
}
}
Here is driver class,
public class Driver {
public static void main(String[] args) {
C c = new C();
c.printClass();
}
}
Output of this will be
In C Class
In A Class
Class B printClass behavior will be ignored in this case.
I am not sure about is this a ideal or good practice to achieve super.super, but still it is working.
Look at this Github project, especially the objectHandle variable. This project shows how to actually and accurately call the grandparent method on a grandchild.
Just in case the link gets broken, here is the code:
import lombok.val;
import org.junit.Assert;
import org.junit.Test;
import java.lang.invoke.*;
/*
Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.
Please don't actually do this... :P
*/
public class ImplLookupTest {
private MethodHandles.Lookup getImplLookup() throws NoSuchFieldException, IllegalAccessException {
val field = MethodHandles.Lookup.class.getDeclaredField("IMPL_LOOKUP");
field.setAccessible(true);
return (MethodHandles.Lookup) field.get(null);
}
#Test
public void test() throws Throwable {
val lookup = getImplLookup();
val baseHandle = lookup.findSpecial(Base.class, "toString",
MethodType.methodType(String.class),
Sub.class);
val objectHandle = lookup.findSpecial(Object.class, "toString",
MethodType.methodType(String.class),
// Must use Base.class here for this reference to call Object's toString
Base.class);
val sub = new Sub();
Assert.assertEquals("Sub", sub.toString());
Assert.assertEquals("Base", baseHandle.invoke(sub));
Assert.assertEquals(toString(sub), objectHandle.invoke(sub));
}
private static String toString(Object o) {
return o.getClass().getName() + "#" + Integer.toHexString(o.hashCode());
}
public class Sub extends Base {
#Override
public String toString() {
return "Sub";
}
}
public class Base {
#Override
public String toString() {
return "Base";
}
}
}
Happy Coding!!!!
I would put the super.super method body in another method, if possible
class SuperSuperClass {
public String toString() {
return DescribeMe();
}
protected String DescribeMe() {
return "I am super super";
}
}
class SuperClass extends SuperSuperClass {
public String toString() {
return "I am super";
}
}
class ChildClass extends SuperClass {
public String toString() {
return DescribeMe();
}
}
Or if you cannot change the super-super class, you can try this:
class SuperSuperClass {
public String toString() {
return "I am super super";
}
}
class SuperClass extends SuperSuperClass {
public String toString() {
return DescribeMe(super.toString());
}
protected String DescribeMe(string fromSuper) {
return "I am super";
}
}
class ChildClass extends SuperClass {
protected String DescribeMe(string fromSuper) {
return fromSuper;
}
}
In both cases, the
new ChildClass().toString();
results to "I am super super"
It would seem to be possible to at least get the class of the superclass's superclass, though not necessarily the instance of it, using reflection; if this might be useful, please consider the Javadoc at http://java.sun.com/j2se/1.5.0/docs/api/java/lang/Class.html#getSuperclass()
public class A {
#Override
public String toString() {
return "A";
}
}
public class B extends A {
#Override
public String toString() {
return "B";
}
}
public class C extends B {
#Override
public String toString() {
return "C";
}
}
public class D extends C {
#Override
public String toString() {
String result = "";
try {
result = this.getClass().getSuperclass().getSuperclass().getSuperclass().newInstance().toString();
} catch (InstantiationException ex) {
Logger.getLogger(D.class.getName()).log(Level.SEVERE, null, ex);
} catch (IllegalAccessException ex) {
Logger.getLogger(D.class.getName()).log(Level.SEVERE, null, ex);
}
return result;
}
}
public class Main {
public static void main(String... args) {
D d = new D();
System.out.println(d);
}
}
run:
A
BUILD SUCCESSFUL (total time: 0 seconds)
I have had situations like these when the architecture is to build common functionality in a common CustomBaseClass which implements on behalf of several derived classes.
However, we need to circumvent common logic for specific method for a specific derived class. In such cases, we must use a super.super.methodX implementation.
We achieve this by introducing a boolean member in the CustomBaseClass, which can be used to selectively defer custom implementation and yield to default framework implementation where desirable.
...
FrameworkBaseClass (....) extends...
{
methodA(...){...}
methodB(...){...}
...
methodX(...)
...
methodN(...){...}
}
/* CustomBaseClass overrides default framework functionality for benefit of several derived classes.*/
CustomBaseClass(...) extends FrameworkBaseClass
{
private boolean skipMethodX=false;
/* implement accessors isSkipMethodX() and setSkipMethodX(boolean)*/
methodA(...){...}
methodB(...){...}
...
methodN(...){...}
methodX(...){
if (isSkipMethodX()) {
setSKipMethodX(false);
super.methodX(...);
return;
}
... //common method logic
}
}
DerivedClass1(...) extends CustomBaseClass
DerivedClass2(...) extends CustomBaseClass
...
DerivedClassN(...) extends CustomBaseClass...
DerivedClassX(...) extends CustomBaseClass...
{
methodX(...){
super.setSKipMethodX(true);
super.methodX(...);
}
}
However, with good architecture principles followed in framework as well as app, we could avoid such situations easily, by using hasA approach, instead of isA approach. But at all times it is not very practical to expect well designed architecture in place, and hence the need to get away from solid design principles and introduce hacks like this.
Just my 2 cents...
IMO, it's a clean way to achieve super.super.sayYourName() behavior in Java.
public class GrandMa {
public void sayYourName(){
System.out.println("Grandma Fedora");
}
}
public class Mama extends GrandMa {
public void sayYourName(boolean lie){
if(lie){
super.sayYourName();
}else {
System.out.println("Mama Stephanida");
}
}
}
public class Daughter extends Mama {
public void sayYourName(boolean lie){
if(lie){
super.sayYourName(lie);
}else {
System.out.println("Little girl Masha");
}
}
}
public class TestDaughter {
public static void main(String[] args){
Daughter d = new Daughter();
System.out.print("Request to lie: d.sayYourName(true) returns ");
d.sayYourName(true);
System.out.print("Request not to lie: d.sayYourName(false) returns ");
d.sayYourName(false);
}
}
Output:
Request to lie: d.sayYourName(true) returns Grandma Fedora
Request not to lie: d.sayYourName(false) returns Little girl Masha
I think this is a problem that breaks the inheritance agreement.
By extending a class you obey / agree its behavior, features
Whilst when calling super.super.method(), you want to break your own obedience agreement.
You just cannot cherry pick from the super class.
However, there may happen situations when you feel the need to call super.super.method() - usually a bad design sign, in your code or in the code you inherit !
If the super and super super classes cannot be refactored (some legacy code), then opt for composition over inheritance.
Encapsulation breaking is when you #Override some methods by breaking the encapsulated code.
The methods designed not to be overridden are marked
final.
In C# you can call a method of any ancestor like this:
public class A
internal virtual void foo()
...
public class B : A
public new void foo()
...
public class C : B
public new void foo() {
(this as A).foo();
}
Also you can do this in Delphi:
type
A=class
procedure foo;
...
B=class(A)
procedure foo; override;
...
C=class(B)
procedure foo; override;
...
A(objC).foo();
But in Java you can do such focus only by some gear. One possible way is:
class A {
int y=10;
void foo(Class X) throws Exception {
if(X!=A.class)
throw new Exception("Incorrect parameter of "+this.getClass().getName()+".foo("+X.getName()+")");
y++;
System.out.printf("A.foo(%s): y=%d\n",X.getName(),y);
}
void foo() throws Exception {
System.out.printf("A.foo()\n");
this.foo(this.getClass());
}
}
class B extends A {
int y=20;
#Override
void foo(Class X) throws Exception {
if(X==B.class) {
y++;
System.out.printf("B.foo(%s): y=%d\n",X.getName(),y);
} else {
System.out.printf("B.foo(%s) calls B.super.foo(%s)\n",X.getName(),X.getName());
super.foo(X);
}
}
}
class C extends B {
int y=30;
#Override
void foo(Class X) throws Exception {
if(X==C.class) {
y++;
System.out.printf("C.foo(%s): y=%d\n",X.getName(),y);
} else {
System.out.printf("C.foo(%s) calls C.super.foo(%s)\n",X.getName(),X.getName());
super.foo(X);
}
}
void DoIt() {
try {
System.out.printf("DoIt: foo():\n");
foo();
Show();
System.out.printf("DoIt: foo(B):\n");
foo(B.class);
Show();
System.out.printf("DoIt: foo(A):\n");
foo(A.class);
Show();
} catch(Exception e) {
//...
}
}
void Show() {
System.out.printf("Show: A.y=%d, B.y=%d, C.y=%d\n\n", ((A)this).y, ((B)this).y, ((C)this).y);
}
}
objC.DoIt() result output:
DoIt: foo():
A.foo()
C.foo(C): y=31
Show: A.y=10, B.y=20, C.y=31
DoIt: foo(B):
C.foo(B) calls C.super.foo(B)
B.foo(B): y=21
Show: A.y=10, B.y=21, C.y=31
DoIt: foo(A):
C.foo(A) calls C.super.foo(A)
B.foo(A) calls B.super.foo(A)
A.foo(A): y=11
Show: A.y=11, B.y=21, C.y=31
It is simply easy to do. For instance:
C subclass of B and B subclass of A. Both of three have method methodName() for example.
public abstract class A {
public void methodName() {
System.out.println("Class A");
}
}
public class B extends A {
public void methodName() {
super.methodName();
System.out.println("Class B");
}
// Will call the super methodName
public void hackSuper() {
super.methodName();
}
}
public class C extends B {
public static void main(String[] args) {
A a = new C();
a.methodName();
}
#Override
public void methodName() {
/*super.methodName();*/
hackSuper();
System.out.println("Class C");
}
}
Run class C Output will be:
Class A
Class C
Instead of output:
Class A
Class B
Class C
If you think you are going to be needing the superclass, you could reference it in a variable for that class. For example:
public class Foo
{
public int getNumber()
{
return 0;
}
}
public class SuperFoo extends Foo
{
public static Foo superClass = new Foo();
public int getNumber()
{
return 1;
}
}
public class UltraFoo extends Foo
{
public static void main(String[] args)
{
System.out.println(new UltraFoo.getNumber());
System.out.println(new SuperFoo().getNumber());
System.out.println(new SuperFoo().superClass.getNumber());
}
public int getNumber()
{
return 2;
}
}
Should print out:
2
1
0
public class SubSubClass extends SubClass {
#Override
public void print() {
super.superPrint();
}
public static void main(String[] args) {
new SubSubClass().print();
}
}
class SuperClass {
public void print() {
System.out.println("Printed in the GrandDad");
}
}
class SubClass extends SuperClass {
public void superPrint() {
super.print();
}
}
Output: Printed in the GrandDad
The keyword super is just a way to invoke the method in the superclass.
In the Java tutorial:https://docs.oracle.com/javase/tutorial/java/IandI/super.html
If your method overrides one of its superclass's methods, you can invoke the overridden method through the use of the keyword super.
Don't believe that it's a reference of the super object!!! No, it's just a keyword to invoke methods in the superclass.
Here is an example:
class Animal {
public void doSth() {
System.out.println(this); // It's a Cat! Not an animal!
System.out.println("Animal do sth.");
}
}
class Cat extends Animal {
public void doSth() {
System.out.println(this);
System.out.println("Cat do sth.");
super.doSth();
}
}
When you call cat.doSth(), the method doSth() in class Animal will print this and it is a cat.
I updated Nexus 5X to Android N, and now when I install the app (debug or release) on it I am getting TransactionTooLargeException on every screen transition that has Bundle in extras. The app is working on all other devices. The old app that is on PlayStore and has mostly same code is working on Nexus 5X.
Is anyone having the same issue?
java.lang.RuntimeException: android.os.TransactionTooLargeException: data parcel size 592196 bytes
at android.app.ActivityThread$StopInfo.run(ActivityThread.java:3752)
at android.os.Handler.handleCallback(Handler.java:751)
at android.os.Handler.dispatchMessage(Handler.java:95)
at android.os.Looper.loop(Looper.java:154)
at android.app.ActivityThread.main(ActivityThread.java:6077)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:865)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:755)
Caused by: android.os.TransactionTooLargeException: data parcel size 592196 bytes
at android.os.BinderProxy.transactNative(Native Method)
at android.os.BinderProxy.transact(Binder.java:615)
at android.app.ActivityManagerProxy.activityStopped(ActivityManagerNative.java:3606)
at android.app.ActivityThread$StopInfo.run(ActivityThread.java:3744)
at android.os.Handler.handleCallback(Handler.java:751)
at android.os.Handler.dispatchMessage(Handler.java:95)
at android.os.Looper.loop(Looper.java:154)
at android.app.ActivityThread.main(ActivityThread.java:6077)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:865)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:755)
Whenever you see TransactionTooLargeException happening when an Activity is in the process of stopping, that means that the Activity was trying to send its saved state Bundles to the system OS for safe keeping for restoration later (after a config change or process death) but that one or more of the Bundles it sent were too large. There is a maximum limit of about 1MB for all such transactions occurring at once and that limit can be reached even if no single Bundle exceeds that limit.
The main culprit here is generally saving too much data inside onSaveInstanceState of either the Activity or any Fragments hosted by the Activity. Typically this happens when saving something particularly large like a Bitmap but it can also happen when sending large quantities of smaller data, like lists of Parcelable objects. The Android team has made very clear on numerous occasions that only small amounts of view-related data should be saved in onSavedInstanceState. However, developers have often saved pages of network data in order to make configuration changes appear as smooth as possible by not having to refetch the same data again. As of Google I/O 2017, the Android team has made clear that the preferred architecture for an Android app saves networking data
in memory so it can be easily reused across configuration changes
to disk so that it can be easily restored after process death and app sessions
Their new ViewModel framework and Room persistence library are meant to help developers fit this pattern. If your problem is with saving too much data in onSaveInstanceState, updating to an architecture like this using those tools should fix your problem.
Personally, before updating to that new pattern I'd like to take my existing apps and just get around the TransactionTooLargeException in the meantime. I wrote a quick library to do just that: https://github.com/livefront/bridge . It uses the same general ideas of restoring state from memory across configuration changes and from disk after process death, rather than sending all that state to the OS via onSaveInstanceState, but requires very minimal changes to your existing code to use. Any strategy that fits those two goals should help you avoid the exception, though, without sacrificing your ability to save state.
On final note here : the only reason you see this on Nougat+ is that originally if the binder transaction limit was exceeded, the process to send the saved state to the OS would fail silently with only this error showing up in Logcat:
!!! FAILED BINDER TRANSACTION !!!
In Nougat, that silent failure was upgraded to a hard crash. To their credit, this is something the development team documented in the release notes for Nougat:
Many platform APIs have now started checking for large payloads being sent across Binder transactions, and the system now rethrows TransactionTooLargeExceptions as RuntimeExceptions, instead of silently logging or suppressing them. One common example is storing too much data in Activity.onSaveInstanceState(), which causes ActivityThread.StopInfo to throw a RuntimeException when your app targets Android 7.0.
In the end, my problem was with things that were being saved onSaveInstance, and not with things that were being sent to the next activity. I removed all saves where I can't control the size of objects (network responses), and now it's working.
Update 2:
Google now provides AndroidX ViewModel which is based on the same technology as retained Fragments but much easier to use. Now ViewModel is a preferred approach.
Update 1:
To preserve big chunks of data, Google is suggesting to do it with Fragment that retains instance. Idea is to create an empty Fragment without a view with all necessary fields, that would otherwise be saved in the Bundle. Add setRetainInstance(true); to Fragment's onCreate method.
And then save data in Fragment on Activity's onDestroy and load them onCreate.
Here is an example of Activity:
public class MyActivity extends Activity {
private DataFragment dataFragment;
#Override
public void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
setContentView(R.layout.main);
// find the retained fragment on activity restarts
FragmentManager fm = getFragmentManager();
dataFragment = (DataFragment) fm.findFragmentByTag(“data”);
// create the fragment and data the first time
if (dataFragment == null) {
// add the fragment
dataFragment = new DataFragment();
fm.beginTransaction().add(dataFragment, “data”).commit();
// load the data from the web
dataFragment.setData(loadMyData());
}
// the data is available in dataFragment.getData()
...
}
#Override
public void onDestroy() {
super.onDestroy();
// store the data in the fragment
dataFragment.setData(collectMyLoadedData());
}
}
An example of Fragment:
public class DataFragment extends Fragment {
// data object we want to retain
private MyDataObject data;
// this method is only called once for this fragment
#Override
public void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
// retain this fragment
setRetainInstance(true);
}
public void setData(MyDataObject data) {
this.data = data;
}
public MyDataObject getData() {
return data;
}
}
More about it, you can read here.
Did a hit and trial, and finally this solved my issue.
Add this to your Activity
#Override
protected void onSaveInstanceState(Bundle oldInstanceState) {
super.onSaveInstanceState(oldInstanceState);
oldInstanceState.clear();
}
The TransactionTooLargeException has been plaguing us for about 4 months now, and we've finally resolved the issue!
What was happening was we are using a FragmentStatePagerAdapter in a ViewPager. The user would page through and create 100+ fragments (its a reading application).
Although we manage the fragments properly in destroyItem(), in Androids
implementation of FragmentStatePagerAdapter there is a bug, where it kept a reference to the following list:
private ArrayList<Fragment.SavedState> mSavedState = new ArrayList<Fragment.SavedState>();
And when the Android's FragmentStatePagerAdapter attempts to save the state, it will call the function
#Override
public Parcelable saveState() {
Bundle state = null;
if (mSavedState.size() > 0) {
state = new Bundle();
Fragment.SavedState[] fss = new Fragment.SavedState[mSavedState.size()];
mSavedState.toArray(fss);
state.putParcelableArray("states", fss);
}
for (int i=0; i<mFragments.size(); i++) {
Fragment f = mFragments.get(i);
if (f != null && f.isAdded()) {
if (state == null) {
state = new Bundle();
}
String key = "f" + i;
mFragmentManager.putFragment(state, key, f);
}
}
return state;
}
As you can see, even if you properly manage the fragments in the FragmentStatePagerAdapter subclass, the base class will still store an Fragment.SavedState for every single fragment ever created. The TransactionTooLargeException would occur when that array was dumped to a parcelableArray and the OS wouldn't like it 100+ items.
Therefore the fix for us was to override the saveState() method and not store anything for "states".
#Override
public Parcelable saveState() {
Bundle bundle = (Bundle) super.saveState();
bundle.putParcelableArray("states", null); // Never maintain any states from the base class, just null it out
return bundle;
}
I face this issue as well on my Nougat devices. My app uses a fragment with a view pager which contains 4 fragments. I passed some large construction arguments to the 4 fragments which caused the problem.
I traced the size of Bundle causing this with the help of TooLargeTool.
Finally, I resolved it using putSerializable on a POJO object which implements Serializable instead of passing a large raw String using putString during fragment initialization. This reduced size of Bundle by half and does not throw the TransactionTooLargeException. Therefore, please make sure you do not pass huge size arguments to Fragment.
P.S. related issue in Google issue tracker: https://issuetracker.google.com/issues/37103380
I face the similar issue. The issue and scenario are little different and I fix it in the following way. Please check the scenario and solution.
Scenario:
I got a weird bug from the customer in the Google Nexus 6P device(7 OS) as my application will crash after 4 hours of working. Later I identify that it's throwing the similar (android.os.TransactionTooLargeException:) exception.
Solution:
The log was not pointing any particular class in the application and later I found that this is happening because of keeping the back stack of fragments. In my case, 4 fragments are added to the back stack repeatedly with the help of an auto screen movement animation. So I override the onBackstackChanged() as mention below.
#Override
public void onBackStackChanged() {
try {
int count = mFragmentMngr.getBackStackEntryCount();
if (count > 0) {
if (count > 30) {
mFragmentMngr.popBackStack(1, FragmentManager.POP_BACK_STACK_INCLUSIVE);
count = mFragmentMngr.getBackStackEntryCount();
}
FragmentManager.BackStackEntry entry = mFragmentMngr.getBackStackEntryAt(count - 1);
mCurrentlyLoadedFragment = Integer.parseInt(entry.getName());
}
} catch (Exception e) {
e.printStackTrace();
}
}
If the stack exceeds the limit, it will automatically pop to initial fragment. I hope somebody will help this answer because the exception and stack trace logs are same. So whenever this issue happens, please check the back stack count, if you are using Fragments and back stack.
In my case, I got that exception inside a fragment because one of its arguments was a very large string that I forgot to delete it (I only used that large string inside the onViewCreated() method). So, to solve this, i simply deleted that argument. In your case, you have to clear or nullify any suspicious field before call onPause().
Activity code
Fragment fragment = new Fragment();
Bundle args = new Bundle();
args.putString("extremely large string", data.getValue());
fragment.setArguments(args);
Fragment code
#Override
public void onViewCreated(View view, Bundle savedInstanceState) {
String largeString = arguments.get("extremely large string");
//Do Something with the large string
arguments.clear() //I forgot to execute this
}
The problem in my app was that I was trying to save too much into the savedInstanceState, the solution was to identify exactly which data should be saved at the right time. Basically look carefully into your onSaveInstanceState to make sure you don't stretch it:
#Override
public void onSaveInstanceState(Bundle savedInstanceState) {
// Save the user's current state
// Check carefully what you're adding into the savedInstanceState before saving it
super.onSaveInstanceState(savedInstanceState);
}
I faced the same issue.
My workaround offloads savedInstanceState to files in cache dir.
I made the following utility class.
package net.cattaka.android.snippets.issue;
import android.content.Context;
import android.content.SharedPreferences;
import android.os.Build;
import android.os.Bundle;
import android.os.Parcel;
import android.os.Parcelable;
import android.support.annotation.NonNull;
import android.support.annotation.Nullable;
import java.io.ByteArrayOutputStream;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;
/**
* To parry BUG of Android N. https://code.google.com/p/android/issues/detail?id=212316
* <p>
* Created by cattaka on 2017/01/12.
*/
public class Issue212316Parrier {
public static final String DEFAULT_NAME = "Issue212316Parrier";
private static final String KEY_STORED_BUNDLE_ID = "net.cattaka.android.snippets.issue.Issue212316Parrier.KEY_STORED_BUNDLE_ID";
private String mName;
private Context mContext;
private String mAppVersionName;
private int mAppVersionCode;
private SharedPreferences mPreferences;
private File mDirForStoredBundle;
public Issue212316Parrier(Context context, String appVersionName, int appVersionCode) {
this(context, appVersionName, appVersionCode, DEFAULT_NAME);
}
public Issue212316Parrier(Context context, String appVersionName, int appVersionCode, String name) {
mName = name;
mContext = context;
mAppVersionName = appVersionName;
mAppVersionCode = appVersionCode;
}
public void initialize() {
mPreferences = mContext.getSharedPreferences(mName, Context.MODE_PRIVATE);
File cacheDir = mContext.getCacheDir();
mDirForStoredBundle = new File(cacheDir, mName);
if (!mDirForStoredBundle.exists()) {
mDirForStoredBundle.mkdirs();
}
long lastStoredBundleId = 1;
boolean needReset = true;
String fingerPrint = (Build.FINGERPRINT != null) ? Build.FINGERPRINT : "";
needReset = !fingerPrint.equals(mPreferences.getString("deviceFingerprint", null))
|| !mAppVersionName.equals(mPreferences.getString("appVersionName", null))
|| (mAppVersionCode != mPreferences.getInt("appVersionCode", 0));
lastStoredBundleId = mPreferences.getLong("lastStoredBundleId", 1);
if (needReset) {
clearDirForStoredBundle();
mPreferences.edit()
.putString("deviceFingerprint", Build.FINGERPRINT)
.putString("appVersionName", mAppVersionName)
.putInt("appVersionCode", mAppVersionCode)
.putLong("lastStoredBundleId", lastStoredBundleId)
.apply();
}
}
/**
* Call this from {#link android.app.Activity#onCreate(Bundle)}, {#link android.app.Activity#onRestoreInstanceState(Bundle)} or {#link android.app.Activity#onPostCreate(Bundle)}
*/
public void restoreSaveInstanceState(#Nullable Bundle savedInstanceState, boolean deleteStoredBundle) {
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.N) {
if (savedInstanceState != null && savedInstanceState.containsKey(KEY_STORED_BUNDLE_ID)) {
long storedBundleId = savedInstanceState.getLong(KEY_STORED_BUNDLE_ID);
File storedBundleFile = new File(mDirForStoredBundle, storedBundleId + ".bin");
Bundle storedBundle = loadBundle(storedBundleFile);
if (storedBundle != null) {
savedInstanceState.putAll(storedBundle);
}
if (deleteStoredBundle && storedBundleFile.exists()) {
storedBundleFile.delete();
}
}
}
}
/**
* Call this from {#link android.app.Activity#onSaveInstanceState(Bundle)}
*/
public void saveInstanceState(Bundle outState) {
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.N) {
if (outState != null) {
long nextStoredBundleId = mPreferences.getLong("lastStoredBundleId", 1) + 1;
mPreferences.edit().putLong("lastStoredBundleId", nextStoredBundleId).apply();
File storedBundleFile = new File(mDirForStoredBundle, nextStoredBundleId + ".bin");
saveBundle(outState, storedBundleFile);
outState.clear();
outState.putLong(KEY_STORED_BUNDLE_ID, nextStoredBundleId);
}
}
}
private void saveBundle(#NonNull Bundle bundle, #NonNull File storedBundleFile) {
byte[] blob = marshall(bundle);
OutputStream out = null;
try {
out = new GZIPOutputStream(new FileOutputStream(storedBundleFile));
out.write(blob);
out.flush();
out.close();
} catch (IOException e) {
// ignore
} finally {
if (out != null) {
try {
out.close();
} catch (IOException e) {
// ignore
}
}
}
}
#Nullable
private Bundle loadBundle(File storedBundleFile) {
byte[] blob = null;
InputStream in = null;
try {
in = new GZIPInputStream(new FileInputStream(storedBundleFile));
ByteArrayOutputStream bout = new ByteArrayOutputStream();
int n;
byte[] buffer = new byte[1024];
while ((n = in.read(buffer)) > -1) {
bout.write(buffer, 0, n); // Don't allow any extra bytes to creep in, final write
}
bout.close();
blob = bout.toByteArray();
} catch (IOException e) {
// ignore
} finally {
if (in != null) {
try {
in.close();
} catch (IOException e) {
// ignore
}
}
}
try {
return (blob != null) ? (Bundle) unmarshall(blob) : null;
} catch (Exception e) {
return null;
}
}
private void clearDirForStoredBundle() {
for (File file : mDirForStoredBundle.listFiles()) {
if (file.isFile() && file.getName().endsWith(".bin")) {
file.delete();
}
}
}
#NonNull
private static <T extends Parcelable> byte[] marshall(#NonNull final T object) {
Parcel p1 = Parcel.obtain();
p1.writeValue(object);
byte[] data = p1.marshall();
p1.recycle();
return data;
}
#SuppressWarnings("unchecked")
#NonNull
private static <T extends Parcelable> T unmarshall(#NonNull byte[] bytes) {
Parcel p2 = Parcel.obtain();
p2.unmarshall(bytes, 0, bytes.length);
p2.setDataPosition(0);
T result = (T) p2.readValue(Issue212316Parrier.class.getClassLoader());
p2.recycle();
return result;
}
}
Full codes: https://github.com/cattaka/AndroidSnippets/pull/37
I worry about that Parcel#marshall should not be used for persistent.
But, I don't have any other idea.
None of the above answers worked for me, the reason of the issue was quite simple as stated by some I was using FragmentStatePagerAdapter and its saveState method saves the state of the fragments, because one of my fragment was quite large , so saving of this fragment leads to this TransactionTooLargeExecption.
I tried overriding the saveState method in my implementation of pager as stated by #IK828, but this couldn't resolve the crash.
My fragment was having an EditText which used to hold very large text, which was the culprit of the issue in my case, so simply in onPause() of fragment, I set the edittext text to empty string.
ie:
#Override
public void onPause() {
edittext.setText("");
}
Now when FragmentStatePagerAdapter will try to saveState, this large chunk of text will not be there to consume bigger part of it, hence resolves the crash.
In your case you need to find whatever is the culprit, it could be an ImageView with some bitmap, a TextView with huge chunk of text or any other high memory consuming view, you need to free it's memory, you may set imageview.setImageResource(null) or similar in onPause() of your fragment.
update : onSaveInstanceState is better place for the purpose before calling super like:
#Override
public void onSaveInstanceState(Bundle outState) {
edittext.setText("");
super.onSaveInstanceState(outState);
}
or as pointed by #Vladimir you can use android:saveEnabled="false" or view.setSaveEnabled(false); on the view or custom view and make sure to set the text back in onResume otherwise it will be empty when Activity resumes.
In my case, I used TooLargeTool to track where the issue was coming from and I found out the android:support:fragments key in the Bundle from my onSaveInstanceState used to reach almost 1mb when the app crashed. So the solution was like:
#Override
public void onSaveInstanceState(Bundle outState) {
super.onSaveInstanceState(outState);
outState.remove("android:support:fragments");
}
By doing that, I avoided to save all fragments' states and kept with other things that need to be saved.
Just Override this method on your activity :
#Override
protected void onSaveInstanceState(Bundle outState) {
// below line to be commented to prevent crash on nougat.
// http://blog.sqisland.com/2016/09/transactiontoolargeexception-crashes-nougat.html
//
//super.onSaveInstanceState(outState);
}
Go to https://code.google.com/p/android/issues/detail?id=212316#makechanges for more info.
As the Android N change the behavior and throw TransactionTooLargeException instead of logging the error.
try {
if (DEBUG_MEMORY_TRIM) Slog.v(TAG, "Reporting activity stopped: " + activity);
ActivityManagerNative.getDefault().activityStopped(
activity.token, state, persistentState, description);
} catch (RemoteException ex) {
if (ex instanceof TransactionTooLargeException
&& activity.packageInfo.getTargetSdkVersion() < Build.VERSION_CODES.N) {
Log.e(TAG, "App sent too much data in instance state, so it was ignored", ex);
return;
}
throw ex.rethrowFromSystemServer();
}
my solution is to hook the ActivityMangerProxy instance and try catch the activityStopped method.
Here is the code:
private boolean hookActivityManagerNative() {
try {
ClassLoader loader = ClassLoader.getSystemClassLoader();
Field singletonField = ReflectUtils.findField(loader.loadClass("android.app.ActivityManagerNative"), "gDefault");
ReflectUtils.ReflectObject singletonObjWrap = ReflectUtils.wrap(singletonField.get(null));
Object realActivityManager = singletonObjWrap.getChildField("mInstance").get();
Object fakeActivityManager = Proxy.newProxyInstance(ClassLoader.getSystemClassLoader(),
new Class[]{loader.loadClass("android.app.IActivityManager")}, new ActivityManagerHook(realActivityManager));
singletonObjWrap.setChildField("mInstance", fakeActivityManager);
return true;
} catch (Throwable e) {
AppHolder.getThirdPartUtils().markException(e);
return false;
}
}
private static class ActivityManagerHook implements InvocationHandler {
private Object origin;
ActivityManagerHook(Object origin) {
this.origin = origin;
}
public Object getOrigin() {
return origin;
}
#Override
public Object invoke(Object proxy, Method method, Object[] args) throws Throwable {
switch (method.getName()) {
//ActivityManagerNative.getDefault().activityStopped(activity.token, state, persistentState, description);
case "activityStopped": {
try {
return method.invoke(getOrigin(), args);
} catch (Exception e) {
e.printStackTrace();
}
return null;
}
}
return method.invoke(getOrigin(), args);
}
}
And the reflect helper class is
public class ReflectUtils {
private static final HashMap<String, Field> fieldCache = new HashMap<>();
private static final HashMap<String, Method> methodCache = new HashMap<>();
public static Field findField(Class<?> clazz, String fieldName) throws Throwable {
String fullFieldName = clazz.getName() + '#' + fieldName;
if (fieldCache.containsKey(fullFieldName)) {
Field field = fieldCache.get(fullFieldName);
if (field == null)
throw new NoSuchFieldError(fullFieldName);
return field;
}
try {
Field field = findFieldRecursiveImpl(clazz, fieldName);
field.setAccessible(true);
fieldCache.put(fullFieldName, field);
return field;
} catch (NoSuchFieldException e) {
fieldCache.put(fullFieldName, null);
throw new NoSuchFieldError(fullFieldName);
}
}
private static Field findFieldRecursiveImpl(Class<?> clazz, String fieldName) throws NoSuchFieldException {
try {
return clazz.getDeclaredField(fieldName);
} catch (NoSuchFieldException e) {
while (true) {
clazz = clazz.getSuperclass();
if (clazz == null || clazz.equals(Object.class))
break;
try {
return clazz.getDeclaredField(fieldName);
} catch (NoSuchFieldException ignored) {
}
}
throw e;
}
}
public static Method findMethodExact(Class<?> clazz, String methodName, Class<?>... parameterTypes) throws Throwable {
String fullMethodName = clazz.getName() + '#' + methodName + getParametersString(parameterTypes) + "#exact";
if (methodCache.containsKey(fullMethodName)) {
Method method = methodCache.get(fullMethodName);
if (method == null)
throw new NoSuchMethodError(fullMethodName);
return method;
}
try {
Method method = clazz.getDeclaredMethod(methodName, parameterTypes);
method.setAccessible(true);
methodCache.put(fullMethodName, method);
return method;
} catch (NoSuchMethodException e) {
methodCache.put(fullMethodName, null);
throw new NoSuchMethodError(fullMethodName);
}
}
/**
* Returns an array of the given classes.
*/
public static Class<?>[] getClassesAsArray(Class<?>... clazzes) {
return clazzes;
}
private static String getParametersString(Class<?>... clazzes) {
StringBuilder sb = new StringBuilder("(");
boolean first = true;
for (Class<?> clazz : clazzes) {
if (first)
first = false;
else
sb.append(",");
if (clazz != null)
sb.append(clazz.getCanonicalName());
else
sb.append("null");
}
sb.append(")");
return sb.toString();
}
/**
* Retrieve classes from an array, where each element might either be a Class
* already, or a String with the full class name.
*/
private static Class<?>[] getParameterClasses(ClassLoader classLoader, Object[] parameterTypes) throws ClassNotFoundException {
Class<?>[] parameterClasses = null;
for (int i = parameterTypes.length - 1; i >= 0; i--) {
Object type = parameterTypes[i];
if (type == null)
throw new ClassNotFoundException("parameter type must not be null", null);
if (parameterClasses == null)
parameterClasses = new Class<?>[i + 1];
if (type instanceof Class)
parameterClasses[i] = (Class<?>) type;
else if (type instanceof String)
parameterClasses[i] = findClass((String) type, classLoader);
else
throw new ClassNotFoundException("parameter type must either be specified as Class or String", null);
}
// if there are no arguments for the method
if (parameterClasses == null)
parameterClasses = new Class<?>[0];
return parameterClasses;
}
public static Class<?> findClass(String className, ClassLoader classLoader) throws ClassNotFoundException {
if (classLoader == null)
classLoader = ClassLoader.getSystemClassLoader();
return classLoader.loadClass(className);
}
public static ReflectObject wrap(Object object) {
return new ReflectObject(object);
}
public static class ReflectObject {
private Object object;
private ReflectObject(Object o) {
this.object = o;
}
public ReflectObject getChildField(String fieldName) throws Throwable {
Object child = ReflectUtils.findField(object.getClass(), fieldName).get(object);
return ReflectUtils.wrap(child);
}
public void setChildField(String fieldName, Object o) throws Throwable {
ReflectUtils.findField(object.getClass(), fieldName).set(object, o);
}
public ReflectObject callMethod(String methodName, Object... args) throws Throwable {
Class<?>[] clazzs = new Class[args.length];
for (int i = 0; i < args.length; i++) {
clazzs[i] = args.getClass();
}
Method method = ReflectUtils.findMethodExact(object.getClass(), methodName, clazzs);
return ReflectUtils.wrap(method.invoke(object, args));
}
public <T> T getAs(Class<T> clazz) {
return (T) object;
}
public <T> T get() {
return (T) object;
}
}
}
Following class I use to create a MediaPlayer object and. This works fine
import android.content.Context;
import android.media.MediaPlayer;
import android.view.View.OnClickListener;
public class PatientAlarm {
private static MediaPlayer md ;
private PatientAlarm() {
}
private PatientAlarm(Context context) {
md = MediaPlayer.create(context, R.raw.patient_alarm);
md.setLooping(true);
}
public static MediaPlayer getMediaPlayer(Context context) {
if (md == null) {
new PatientAlarm(context);
}
return md;
}
public static void startMediaPlayer(Context context) {
if (md == null) {
new PatientAlarm(context);
md.start();
} else {
if (!md.isPlaying()) {
new PatientAlarm(context);
md.start();
}
}
}
public static void stopMediaPlayer(OnClickListener onClickListener) {
if (md == null) {
} else {
if (md.isPlaying()) {
md.stop();
}
}
}
}
But the problem is Klocwork analyzer says
NPE.COND: Null pointer dereference of 'md' where null comes from condition : Java : PatientAlarm.java : Critical (1) : Analyze : Local
That occurs in below line
md.start();
I have no idea what it does mean. before start md, I check whether it is null.
Please guide me to make my code more reliable and how to solve this issue,
You totally messed up the concept of Object Oriented Programming.
You defined PatientAlarm class to incapsulate all logic of working with MediaPlayer. It's OK.
But why did you make all methods static? Why new PatientAlarm() called inside these methods and created instance even not assigned to some variable or field that makes it target for garbage collection?
This is example of how you should do it:
public class PatientAlarm {
private MediaPlayer md;
public PatientAlarm(Context context) {
md = MediaPlayer.create(context, R.raw.patient_alarm);
md.setLooping(true);
md.prepare();
}
public void start() {
if (!md.isPlaying()) {
md.start();
}
}
public void stop() {
if (md.isPlaying()) {
md.stop();
}
}
}
Adapt it to your case and read more about using MediaPlayer class.
which md.start(); line? there are 2... anyway your code is totally vulnerable for concurrency issues, you may read about locks to prevent it, but I would start form refactoring, before you get into advanced stuff. Creating a new object just to initialize a static member of the class is probably the most ugly approach for singleton I have ever seen.. If you really need to, use static method to do it for you, at least there wont be a new object for GC to collect. But in fact the use of static object in your example seems to be a miss first of all. Try rewriting it without static fields or methods and most likely your code will be simpler and more reliable.
I've been experimenting with using flow and mortar as an alternative architecture for our Android apps. I've been working on an app which at the minute is only a single phone layout, but I was wondering how the flow and mortar architecture might work if you want to have a different layout for tablets. Master details might be the simplest example, but there are obviously other examples.
I have a few ideas how this could work, but I wanted to know what the square developers might have thought about already around this topic.
We're still working on a canonical answer for this, but the basic idea is that you let the resource system change what views you're showing in which situation. So your activity sets its content view to, say, R.layout.root_view. The tablet version of that layout (we put it in res/layout-sw600dp) can be tied to different views, which might inject different presenters, and so on.
For cases where you need to make a runtime decision, define a boolean resource in values/bools .xml
<?xml version="1.0" encoding="utf-8"?>
<resources>
<bool name="show_tablet_ui">false</bool>
</resources>
and values-sw600dp/bools.xml
<?xml version="1.0" encoding="utf-8"?>
<resources>
<bool name="show_tablet_ui">true</bool>
</resources>
Expose it to the rest of the app via dagger. Use this binding annotation:
/**
* Whether we should show a tablet UI.
*/
#Retention(RUNTIME) #Qualifier
public #interface ShowTabletUi {
int ID = R.bool.show_tablet_ui;
}
and a provider method like:
/**
* Singleton because there's no reason to read it from resources again,
* it won't change.
*/
#Provides #ShowTabletUi #Singleton boolean showTabletUi(Resources resources) {
return resources.getBoolean(ShowTabletUi.ID);
}
But wait there's more! Suppose you want to have a single screen / blueprint definition that manufactures different modules for different form factors. We've started using an annotation scheme to simplify that kind of thing. Instead of making our screen classes all implement BluePrint, we've started using some annotations to declare their interface class. In that world here's how a screen can selectively choose what modules to use for tablet or mobile.
#Layout(R.layout.some_view) #WithModuleFactory(SomeScreen.ModuleFactory.class)
public class SomeScreen {
public static class ModuleFactory extends ResponsiveModuleFactory<HomeScreen> {
#Override protected Object createTabletModule(HomeScreen screen) {
return new TabletModule();
}
#Override protected Object createMobileModule(HomeScreen screen) {
return new MobileModule();
}
}
Magic, right? Here's what's behind the curtain. First, a ModuleFactory is some static class that is given access to the screen and the resources and spits out a dagger module.
public abstract class ModuleFactory<T> {
final Blueprint createBlueprint(final Resources resources, final MortarScreen screen) {
return new Blueprint() {
#Override public String getMortarScopeName() {
return screen.getName();
}
#Override public Object getDaggerModule() {
return ModuleFactory.this.createDaggerModule(resources, (T) screen);
}
};
}
protected abstract Object createDaggerModule(Resources resources, T screen);
}
Our trixie ResponsiveModuleFactory subclass looks like this. (Remember how ShowTabletUi.java defined the resource id as a constant? This is why.)
public abstract class ResponsiveModuleFactory<T> extends ModuleFactory<T> {
#Override protected final Object createDaggerModule(Resources resources, T screen) {
boolean showTabletUi = resources.getBoolean(ShowTabletUi.ID);
return showTabletUi ? createTabletModule(screen) : createMobileModule(screen);
}
protected abstract Object createTabletModule(T screen);
protected abstract Object createMobileModule(T screen);
}
To make all this go, we have a ScreenScoper class (below). In the Mortar sample code, you'd make the ScreenConductor use one of these to create and destroy scopes. Sooner or later (soon I hope) Mortar and/or its samples will be updated to include this stuff.
package mortar;
import android.content.Context;
import android.content.res.Resources;
import com.squareup.util.Objects;
import dagger.Module;
import java.lang.reflect.Constructor;
import java.lang.reflect.InvocationTargetException;
import java.util.LinkedHashMap;
import java.util.Map;
import static java.lang.String.format;
/**
* Creates {#link MortarScope}s for screens that may be annotated with {#link WithModuleFactory},
* {#link WithModule} or {#link Module}.
*/
public class ScreenScoper {
private static final ModuleFactory NO_FACTORY = new ModuleFactory() {
#Override protected Object createDaggerModule(Resources resources, Object screen) {
throw new UnsupportedOperationException();
}
};
private final Map<Class, ModuleFactory> moduleFactoryCache = new LinkedHashMap<>();
public MortarScope getScreenScope(Context context, final MortarScreen screen) {
MortarScope parentScope = Mortar.getScope(context);
return getScreenScope(context.getResources(), parentScope, screen);
}
/**
* Finds or creates the scope for the given screen, honoring its optoinal {#link
* WithModuleFactory} or {#link WithModule} annotation. Note the scopes are also created
* for unannotated screens.
*/
public MortarScope getScreenScope(Resources resources, MortarScope parentScope,
final MortarScreen screen) {
ModuleFactory moduleFactory = getModuleFactory(screen);
MortarScope childScope;
if (moduleFactory != NO_FACTORY) {
Blueprint blueprint = moduleFactory.createBlueprint(resources, screen);
childScope = parentScope.requireChild(blueprint);
} else {
// We need every screen to have a scope, so that anything it injects is scoped. We need
// this even if the screen doesn't declare a module, because Dagger allows injection of
// objects that are annotated even if they don't appear in a module.
Blueprint blueprint = new Blueprint() {
#Override public String getMortarScopeName() {
return screen.getName();
}
#Override public Object getDaggerModule() {
return null;
}
};
childScope = parentScope.requireChild(blueprint);
}
return childScope;
}
private ModuleFactory getModuleFactory(MortarScreen screen) {
Class<?> screenType = Objects.getClass(screen);
ModuleFactory moduleFactory = moduleFactoryCache.get(screenType);
if (moduleFactory != null) return moduleFactory;
WithModule withModule = screenType.getAnnotation(WithModule.class);
if (withModule != null) {
Class<?> moduleClass = withModule.value();
Constructor<?>[] constructors = moduleClass.getDeclaredConstructors();
if (constructors.length != 1) {
throw new IllegalArgumentException(
format("Module %s for screen %s should have exactly one public constructor",
moduleClass.getName(), screen.getName()));
}
Constructor constructor = constructors[0];
Class[] parameters = constructor.getParameterTypes();
if (parameters.length > 1) {
throw new IllegalArgumentException(
format("Module %s for screen %s should have 0 or 1 parameter", moduleClass.getName(),
screen.getName()));
}
Class screenParameter;
if (parameters.length == 1) {
screenParameter = parameters[0];
if (!screenParameter.isInstance(screen)) {
throw new IllegalArgumentException(format("Module %s for screen %s should have a "
+ "constructor parameter that is a super class of %s", moduleClass.getName(),
screen.getName(), screen.getClass().getName()));
}
} else {
screenParameter = null;
}
try {
if (screenParameter == null) {
moduleFactory = new NoArgsFactory(constructor);
} else {
moduleFactory = new SingleArgFactory(constructor);
}
} catch (Exception e) {
throw new RuntimeException(
format("Failed to instantiate module %s for screen %s", moduleClass.getName(),
screen.getName()), e);
}
}
if (moduleFactory == null) {
WithModuleFactory withModuleFactory = screenType.getAnnotation(WithModuleFactory.class);
if (withModuleFactory != null) {
Class<? extends ModuleFactory> mfClass = withModuleFactory.value();
try {
moduleFactory = mfClass.newInstance();
} catch (Exception e) {
throw new RuntimeException(format("Failed to instantiate module factory %s for screen %s",
withModuleFactory.value().getName(), screen.getName()), e);
}
}
}
if (moduleFactory == null) moduleFactory = NO_FACTORY;
moduleFactoryCache.put(screenType, moduleFactory);
return moduleFactory;
}
private static class NoArgsFactory extends ModuleFactory<Object> {
final Constructor moduleConstructor;
private NoArgsFactory(Constructor moduleConstructor) {
this.moduleConstructor = moduleConstructor;
}
#Override protected Object createDaggerModule(Resources resources, Object ignored) {
try {
return moduleConstructor.newInstance();
} catch (InstantiationException | IllegalAccessException | InvocationTargetException e) {
throw new RuntimeException(e);
}
}
}
private static class SingleArgFactory extends ModuleFactory {
final Constructor moduleConstructor;
public SingleArgFactory(Constructor moduleConstructor) {
this.moduleConstructor = moduleConstructor;
}
#Override protected Object createDaggerModule(Resources resources, Object screen) {
try {
return moduleConstructor.newInstance(screen);
} catch (InstantiationException | IllegalAccessException | InvocationTargetException e) {
throw new RuntimeException(e);
}
}
}
}
I'm working on an Android app and we're investigating memory use.
Looking at a heap dump from hprof, we're seeing nearly 2M (22% of our heap) being used in a static cache in JarURLConnectionImpl:
Looking at the source code for JarURLConnectionImpl, it appears that entries are added to the static jarCache variable, but never removed.
If it's true that they're never removed, that strikes me as a potential memory leak.
Is this a leak? Is there a fix or workaround?
Here's an ugly workaround:
private static HashMap<URL,JarFile> jarCache;
static {
try {
Class<?> jarURLConnectionImplClass = Class.forName("org.apache.harmony.luni.internal.net.www.protocol.jar.JarURLConnectionImpl");
final Field jarCacheField = jarURLConnectionImplClass.getDeclaredField("jarCache");
jarCacheField.setAccessible(true);
//noinspection unchecked
jarCache = (HashMap<URL, JarFile>) jarCacheField.get(null);
} catch(Exception e) {
// ignored
}
}
Then, periodically run the following:
// HACK http://stackoverflow.com/questions/14610350/android-memory-leak-in-apache-harmonys-jarurlconnectionimpl
if( jarCache!=null ) {
try {
for (
final Iterator<Map.Entry<URL, JarFile>> iterator = jarCache.entrySet().iterator(); iterator.hasNext(); ) {
final Map.Entry<URL, JarFile> e = iterator.next();
final URL url = e.getKey();
if (Strings.toString(url).endsWith(".apk")) {
Log.i(TAG,"Removing static hashmap entry for " + url);
try {
final JarFile jarFile = e.getValue();
jarFile.close();
iterator.remove();
} catch( Exception f ) {
Log.e(TAG,"Error removing hashmap entry for "+ url,f);
}
}
}
} catch( Exception e ) {
// ignored
}
}
I run it on activity creation so it gets executed every time one of my activities is created. The ugly hashmap entry doesn't seem to get recreated all that often, but it DOES seem to reappear occasionally, so it's not sufficient to just run this code once.
This is definitely a nasty memory leak. I've opened an issue for it since no one else seems to have reported it.
Thanks for the "ugly workaround" emmby, that was helpful. A safer approach, although potentially with a performance impact, is to disable URLConnection caching altogether. Since the URLConnection.defaultUseCaches flag is static and, as you might guess, is the default for each instance's useCaches flag, you can just set this to false and no more instances will cache their connections. This will affect all implementations of URLConnection, so it may have farther-ranging effects than desired, but I think it's a reasonable trade-off.
You can just create a simple class like this and instantiate it very early in your app's onCreate():
public class URLConnectionNoCache extends URLConnection {
protected URLConnectionNoCache(URL url) {
super(url);
setDefaultUseCaches(false);
}
public void connect() throws IOException {
}
}
The interesting thing is that since this occurs after your app is loaded and run, the system libs should already be cached, and this will only prevent further caching, so this probably gives the best possible trade-off: not caching your apk while allowing the performance benefits of caching the system jars.
Before doing this I did modify emmby's solution a bit to make it a standalone class that creates a background thread to periodically clear the cache. And I restricted it to just clear the app's apk, though that can be relaxed if desired. The thing to worry about here is that you're modifying the objects while they may be in use, which is generally not a good thing. If you do want to go this route you just need to call the start() method with a context, e.g. in your app's onCreate().
package com.example;
import java.lang.reflect.Field;
import java.net.URL;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.jar.JarFile;
import java.util.regex.Pattern;
import android.content.Context;
// hack to remove memory leak in JarURLConnectionImpl
// from http://stackoverflow.com/questions/14610350/android-memory-leak-in-apache-harmonys-jarurlconnectionimpl
public class JarURLMonitor {
private static JarURLMonitor instance;
private Pattern pat;
private Field jarCacheField;
public volatile boolean stop;
private static final long CHECK_INTERVAL = 60 * 1000;
public static synchronized void start(Context context) {
if (instance == null) {
instance = new JarURLMonitor(context);
}
}
public static synchronized void stop() {
if (instance != null) {
instance.stop = true;
}
}
private JarURLMonitor(Context context) {
// get jar cache field
try {
final Class<?> cls = Class.forName("libcore.net.url.JarURLConnectionImpl");
jarCacheField = cls.getDeclaredField("jarCache");
jarCacheField.setAccessible(true);
}
catch (Exception e) {
// log
}
if (jarCacheField != null) {
// create pattern that matches our package: e.g. /data/app/<pkgname>-1.apk
pat = Pattern.compile("^.*/" + context.getPackageName() + "-.*\\.apk$");
// start background thread to check it
new Thread("JarURLMonitor") {
#Override
public void run() {
try {
while (!stop) {
checkJarCache();
Thread.sleep(CHECK_INTERVAL);
}
}
catch (Exception e) {
// log
}
}
}.start();
}
}
private void checkJarCache() throws Exception {
#SuppressWarnings("unchecked")
final HashMap<URL, JarFile> jarCache = (HashMap<URL, JarFile>)jarCacheField.get(null);
final Iterator<Map.Entry<URL, JarFile>> iterator = jarCache.entrySet().iterator();
while (iterator.hasNext()) {
final Map.Entry<URL, JarFile> entry = iterator.next();
final JarFile jarFile = entry.getValue();
final String file = jarFile.getName();
if (pat.matcher(file).matches()) {
try {
jarFile.close();
iterator.remove();
}
catch (Exception e) {
// log
}
}
}
}
}