DialogSurfaceManager implementation.

media/base/android/java/src/org/chromium/media/DialogSurfaceCore.java

+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+package org.chromium.media;
+
+import android.annotation.SuppressLint;
+import android.app.Dialog;
+import android.content.Context;
+import android.os.IBinder;
+import android.view.Gravity;
+import android.view.SurfaceHolder;
+import android.view.Window;
+import android.view.WindowManager;
+
+import org.chromium.base.Log;
+
+/**
+ * Core class for control of a single DialogSurface instance.  Everything runs
+ * on a single thread, that's probably not the UI thread.
+ *
+ * Note that this does not implement IDialogSurface; we assume that, and the
+ * associated thread-hopping, is handled elsewhere.
+ */
+class DialogSurfaceCore {
+    private static final String TAG = "cr_media";

[boliu, 2017/01/04 01:48:44]

generally, this is just the name of the class, the "cr_" art is added
automatically by base.Log I think

[liberato (no reviews please), 2017/01/11 22:17:57]

not sure.  i copied the pattern from MediaCodecBridge.

personally, i prefer the class name too, so i've changed it.  shortened, since i
think there's a low length limit before it truncates.  and elsewhere.

[boliu, 2017/01/12 20:24:19]

fwiw, it's dynamic:
https://cs.chromium.org/chromium/src/base/android/java/src/org/chromium/base/...

[liberato (no reviews please), 2017/02/03 21:28:32]

Acknowledged.

+
+    /**
+     * Call back into native with a message about our state.  This can be called
+     * on any thread.  It's okay if it refers to a native object that no longer
+     * exists; we can't really avoid it.  The native side handles it.
+     *
+     * The callback cannot recursively call back into us; the IDialogSurface
+     * implementation must prevent that.
+     */
+    private IDialogSurfaceCallback mClientCallback;
+
+    // Callback operations.
+    // These must match dialog_surface.h .
+    private static final int OP_CREATED = 0;
+    private static final int OP_DESTROYED = 1;
+
+    // When initialized via Init, we'll create mDialog.  We'll clear it when
+    // we send SURFACE_DESTROYED to the client.  In general, when this is null,
+    // either we haven't been initialized yet, or we've been torn down.  It
+    // shouldn't be the case that anything calls methods after construction but
+    // before Init, though.
+    private Dialog mDialog;
+
+    private Callbacks mDialogCallbacks;
+    private IBinder mToken;
+
+    // Initial position and size.
+    private int mInitialX;
+    private int mInitialY;
+    private int mInitialWidth;
+    private int mInitialHeight;
+
+    /**
+     * Construction may be called from a random thread, for simplicity.  Call
+     * Init from the proper thread before doing anything else.
+     */
+    public DialogSurfaceCore() {}
+
+    /**
+     * Finish init on the proper thread.  We'll use this thread for the Dialog
+     * Looper thread.
+     * @param context Context that we use.
+     * @param callback callback object to notify about state changes.
+     * @param x initial x position in chrome compositor (not screen) coords.
+     * @param y initial y position in chrome compositor (not screen) coords.
+     * @param width initial width.
+     * @param height initial height.
+     */
+    public void initialize(
+            Context context, IDialogSurfaceCallback callback, int x, int y, int width, int height) {
+        mClientCallback = callback;
+        mInitialX = x;
+        mInitialY = y;
+        mInitialWidth = width;
+        mInitialHeight = height;
+
+        // Create the dialog, but don't lay it out or show it yet.  We'll do
+        // that when we get a window token.
+        mDialog = new Dialog(context, android.R.style.Theme_NoDisplay);
+        mDialog.requestWindowFeature(Window.FEATURE_NO_TITLE);
+        mDialog.setCancelable(false);
+    }
+
+    /**
+     * Release the underlying surface, and generally clean up, in response to
+     * the client releasing the IDialogSurface.
+     */
+    public void release() {
+        // If we've not released the dialog yet, then do so.
+        if (mDialog != null) {
+            // TODO(liberato): is this safe if we haven't shown the dialog yet?
+            mDialog.dismiss();
+            mDialog = null;
+            mDialogCallbacks = null;
+        }
+    }
+
+    /**
+     * Layout the DialogSurface.  If we don't have a token, then we ignore it,
+     * since a well-behaved client shouldn't call us before OP_CREATED anyway.
+     */
+    @SuppressLint("RtlHardcoded")
+    public void layoutSurface(final int x, final int y, final int width, final int height) {
+        if (mDialog == null || mToken == null) return;
+
+        // Rather than using getAttributes, we just create them from scratch.
+        // The default dialog attributes aren't what we want.
+        WindowManager.LayoutParams layoutParams = new WindowManager.LayoutParams();

[boliu, 2017/01/04 01:48:44]

ditto about allocating java objects in calls that could happen every frame

[liberato (no reviews please), 2017/01/11 22:17:57]

Done.

+
+        // TODO(liberato): adjust for CompositorView screen location here if we
+        // want to support non-full screen use cases.
+
+        layoutParams.x = x;
+        layoutParams.y = y;
+        layoutParams.width = width;
+        layoutParams.height = height;
+        layoutParams.token = mToken;
+
+        // NOTE: we really do want LEFT here, since we're dealing in compositor
+        // coordinates.  Those are always from the left.
+        layoutParams.gravity = Gravity.TOP | Gravity.LEFT;
+
+        // Use a media surface, which is what SurfaceView uses by default.  For
+        // debugging overlay drawing, consider using TYPE_APPLICATION_PANEL to
+        // move the dialog over the CompositorView.
+        layoutParams.type = WindowManager.LayoutParams.TYPE_APPLICATION_MEDIA;
+
+        layoutParams.flags = WindowManager.LayoutParams.FLAG_NOT_TOUCH_MODAL
+                | WindowManager.LayoutParams.FLAG_NOT_TOUCHABLE
+                | WindowManager.LayoutParams.FLAG_NOT_FOCUSABLE
+                | WindowManager.LayoutParams.FLAG_LAYOUT_NO_LIMITS;
+
+        // Don't set FLAG_SCALED.  in addition to not being sure what it does
+        // (SV uses it), it also causes a crash in WindowManager when we hide
+        // (not dismiss), navigate, and/or exit the app without hide/dismiss.
+        // There's a missing null check in WindowManagerService.java@3170
+        // on M MR2.  To repro, change dimiss() to hide(), bring up a SV, and
+        // navigate away or press home.
+
+        // Turn off the position animation, so that it doesn't animate from one
+        // position to the next.  Ignore errors.
+        try {
+            int currentFlags =
+                    (Integer) layoutParams.getClass().getField("privateFlags").get(layoutParams);

[boliu, 2017/01/04 01:48:44]

err, reflection? hard coded values?

[liberato (no reviews please), 2017/01/11 22:17:57]

PRIVATE_FLAG_NO_MOVE_ANIMATION -- there's no other way to do this,
unfortunately.  otherwise, the thing slides around the screen when it's resized.

[boliu, 2017/01/12 20:24:19]

The bar for using private APIs is very high, ie it's almost never allowed.

It doesn't affect fullscreen, right? Also is it that bad? You can never fully
synchronize surface updates or placements for inline video, so how much worse is
it?

If you really want to do this, then I want to see communication with the android
team to expose these things in the next version of android.

[liberato (no reviews please), 2017/02/03 21:28:32]

i'd really like to do this.

we can chat offline about how it will work in O and later.

+            layoutParams.getClass()
+                    .getField("privateFlags")
+                    .set(layoutParams, currentFlags | 0x00000040);
+            // It would be nice to just catch Exception, but findbugs doesn't
+            // allow it.  If we cannot set the flag, then that's okay too.
+        } catch (NoSuchFieldException e) {
+        } catch (NullPointerException e) {
+        } catch (SecurityException e) {
+        } catch (IllegalAccessException e) {
+        } catch (IllegalArgumentException e) {
+        } catch (ExceptionInInitializerError e) {
+        }
+
+        mDialog.getWindow().setAttributes(layoutParams);
+    }
+
+    /**
+     * Callbacks for finding out about the Dialog's Surface.
+     * These happen on the looper thread.
+     */
+    private class Callbacks implements SurfaceHolder.Callback2 {
+        @Override
+        public void surfaceChanged(SurfaceHolder holder, int format, int width, int height) {}
+
+        @Override
+        public void surfaceCreated(SurfaceHolder holder) {
+            // Make sure that we haven't torn down the dialog yet.
+            if (mDialog == null) return;
+
+            try {
+                mClientCallback.onCreated(holder.getSurface());
+            } catch (Exception e) {
+                Log.e(TAG, "surfaceCreated: callback failed: " + e);
+                enterDestroyedState();
+            }
+        }
+
+        @Override
+        public void surfaceDestroyed(SurfaceHolder holder) {
+            if (mDialog == null) return;
+
+            enterDestroyedState();
+        }
+
+        @Override
+        public void surfaceRedrawNeeded(SurfaceHolder holder) {}
+    }
+
+    /**
+     * Notify the client about surface destruction, and clean up.  Multiple
+     * calls to us do nothing.
+     */
+    private void enterDestroyedState() {
+        if (mDialog == null) return;
+
+        try {
+            mClientCallback.onDestroyed();
+        } catch (Exception e) {
+            Log.e(TAG, "enterDestroyedState: callback failed: " + e);

[boliu, 2017/01/04 01:48:44]

all logs probably should be .d so they get stripped out of release builds (I
don't think there's the equivalent of DLOG/VLOG for java yet)

[liberato (no reviews please), 2017/01/11 22:17:57]

Done.

+            release();
+        }
+
+        // Dismiss the dialog, if needed
+        // TODO(liberato): is it safe to call dismiss without show?  we don't
+        // know that we've shown the dialog yet.  maybe if mToken?  isShowing?
+        mDialog.dismiss();
+        mDialog = null;
+        mDialogCallbacks = null;
+        mToken = null;
+    }
+
+    public void onWindowToken(IBinder token) {
+        if (mDialog == null) return;
+
+        if (token == null || (mToken != null && token != mToken)) {
+            // We've lost the token, if we had one, or we got a new one.
+            // Notify the client.
+            enterDestroyedState();
+            return;
+        }
+
+        if (mToken == token) {
+            // Same token, do nothing.

[boliu, 2017/01/04 01:48:44]

haven't read the window token stuff yet.. but is this expected to happen?

[liberato (no reviews please), 2017/01/11 22:17:57]

i don't think that it should, but it seemed harmless.

[boliu, 2017/01/12 20:24:19]

but this is silencing a potential problem in the future. If this should be an
assert, then assert. Java asserts work like DCHECks now

+            return;
+        }
+
+        // We have a token, so layout the dialog.
+        mToken = token;
+        layoutSurface(mInitialX, mInitialY, mInitialWidth, mInitialHeight);

[boliu, 2017/01/04 01:48:44]

what if an layout happens before window token is received? right now that layout
is just dropped, is that correct?

[liberato (no reviews please), 2017/01/11 22:17:57]

it's okay, since the application shouldn't be positioning the surface before it
gets notification that it's been created.  however, now that you mention it,
it's just as easy to make it stick around.

+        mDialogCallbacks = new Callbacks();
+        mDialog.getWindow().takeSurface(mDialogCallbacks);
+        mDialog.show();
+
+        // We don't notify the client here.  We'll wait until the Android
+        // Surface is created.
+    }
+}