Com.android.internal.policy.impl.PhoneLayoutInflater sometimes stays in memory (hprof dumps)

Question

I check the memory, trying to find possible memory leaks through hprof dumps.

I find that sometimes, when I leave the activity through the back button (which would complete the action), the activity will still be remembered, but it will have only one GC root, which does not seem very “strong” though.

This is my activity stream / the way I click and test:

A, B, C - actions.

1) A → B → (back) A

2) dump hprof with the following result:

B is still in memory , the only elements in the GC root of activity B are:

com.myapp.android.activity.directory.B

• mContext com.android.internal.policy.impl.PhoneLayoutInflater

  • mLayoutInflater of android.app.ContextImpl [Stack Local]

    • [local variable] java.lang.Thread [Thread] "main"

• mOuterContext android.app.ContextImpl [Stack Local]

  • [local variable] java.lang.Thread [Thread] "main"

(The theme "main" seems to be a UI thread)

continue from A:

3) A → C → (back) A

4) dump hprof with the following result (as expected):

B is no longer in memory , C is no longer in memory, but only A

Now my question is: where does this PhoneLayoutInflater come from / why does it stay in memory when I return from B to A, but it will disappear after switching to C and return to A.

Obviously, PhoneLayoutInflater is designed to inflate ideas, I know about this purpose. I just don’t understand why it will be stored in memory through the GC root from the main user interface thread.

When I check the GC roots from the above

[local variable] java.lang.Thread [Thread] "main

he will have the following:

• mUiThread of com.myapp.android.activity.main. A [Stack Local]
  • ....
    • this $ 0 of android.view.inputmethod.InputMethodManager $ ControlledInputConnectionWrapper [JNI Global]

What I call actions B and C from A is done using regular startActivity(intent)

Why is the main thread of user interface A somehow connected and refers to activity B?

Answer 1

Here is what I did in the end, but I still don't like the fact that I use reflection to fix this. Any??

 public static boolean ReleaseActivityContext(Context context) { LayoutInflater inflater = (LayoutInflater)context.getSystemService(Context.LAYOUT_INFLATER_SERVICE); Class<?> i = inflater.getClass(); try { Field f = i.getSuperclass().getDeclaredField("mContext"); f.setAccessible(true); // prevent IllegalAccessException f.set(inflater, null); // can cause IllegalAccessException return true; } catch (Exception e) { e.printStackTrace(); } return false; } 

Answer 2

I had the same problem, and I used the Eclipse MAT to analyze hprof, I spent more than 8 hours, and finally I found that there are some simple rules that should be followed.

Do not leave long-term references to context-activity (a link to an activity should have the same life cycle as the activity itself)

Try using a contextual application instead of contextual activity.

Avoid non-static inner classes in if you do not control their life cycle, use a static inner class and make a weak reference to the activity inside.

Follow these steps:

Then click ...

you can see the reason for the stay of objects in memory

Now review your code and see why there are some links when there is no need for them to exist. then click on the back button to cancel these links and viola .. the world will look beautiful again :)