getSupportFragmentManager() cause java.lang.IllegalStateException: Can not perform this action after onSaveInstanceState

Question

I was trying to figure out why:

getSupportFragmentManager().beginTransaction().commit();

fails, with

java.lang.IllegalStateException: Can not perform this action after onSaveInstanceState

in a very basic FragmentActivity class.

So here was my use-case (this will be some pseudo-code and not a complete example, sorry): I had one FragmentActivity with an internal AsyncTask class. Roughly something like this:

public class HelloWorld extends FragmentActivity {
    showFragment(Fragment fragment, String name) {
        getSupportFragmentManager().beginTransaction().replace(R.id.fragmentContainer, fragment, name).commit();
    }

    private class SlowFragmentShow extends AsyncTask<Context, String, Void> {
        protected Void doInBackground(Context... contexts) {
            try {
                Thread.sleep(10000);
            } catch (InterruptedException e) {
                /* meh */
            }
        }

        protected void onPostExecute(Void nothing) {
            showFragment(new MyFragment(), "myFragment");
        }
    }
}

Or basically, 10 seconds after starting the app, it would show another fragment. Sounds simple, right? This seemed to work well too, until I decided to rotate the phone. When I did that, the app would crash upon calling "getSupportFragmentManager()..." with the "Can not perform this action...".

Answer 1

After lots of debugging, it turned out that when SlowFragmentShow.onPostExecute() was called, which called showFragment(), which in turn called getSupportFragmentManager(), I received a FragmentManager that was in an IllegalState (so arguably the exception I got was correct). I'm still not sure why getSupportFragmentManager() would ever return an object in such a limbo state, but it did, and I needed to somehow get access to the "correct" FragmentManager. So to cut to the chase, I stored the FragmentManager as a static variable in my HelloWorld FragmentActivity, which I updated when HelloWorld.onStart() was called:

public class HelloWorld extends FragmentActivity {
    private static FragmentManager fragmentManager;

    public void onStart() {
        fragmentManager = getSupportFragmentManager();
        /* more code here */
    }

    showFragment(Fragment fragment, String name) {
        fragmentManager.beginTransaction().replace(R.id.fragmentContainer, fragment, name).commit();
    }

    private class SlowFragmentShow extends AsyncTask<Context, String, Void> {
        protected Void doInBackground(Context... contexts) {
            try {
                Thread.sleep(10000);
            } catch (InterruptedException e) {
                /* meh */
            }
        }

        protected void onPostExecute(Void nothing) {
            showFragment(new MyFragment(), "myFragment");
        }
    }
}

And well, that pretty much fixed it. Now I could rotate the phone to my hearts desire, the fragment would still be shown when the AsyncTask was done.

In retrospect, it really seems a bit "oh, of course!", but the design decisions behind Android feels quite "alien" and unusual. I seem to end up with try-catch(Exception) around pretty much half the code just to prevent it from crashing on a non-fatal error (such as failing to update a text field), and a lot of static variables that needs to be updated upon onStart() because that seems like the only sane way you can reference Android objects without them being in an IllegalState.