This is expected for the current release as it is an artifact of the current Trackpad mode impl.  I do want to improve that experience so I will use this bug to track it.

An important note is that this does not affect the remote cursor, it is still constrained to the actual desktop.  This is only a visual representation of the desired cursor position.

Since the definition of image bounds is a set of allowed cursor positions in the image space, I don't quite understand why the image bounds is shift up and get negative y-start value (e.g. (0.0, -140.70222, 2560.0, 1528.5378)). Shouldn't it always be (0, 0, image_width, image_height)?

I think we may be able to mathematically simplify the logic. I'll let you know when you are back :)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/95ccb26c84a537f696b374097f7050d472fba289

commit 95ccb26c84a537f696b374097f7050d472fba289
Author: joedow <joedow@chromium.org>
Date: Tue Oct 25 20:53:58 2016

Updating the calculations used for image overpanning

This CL simplifies the calculations used for panning the remote desktop
image out from under the System UI.  It also constrains the cursor to
the image coordinates vs. the previous method which did not.

The old approach I used was to allow overpanning the remote desktop
image by adjusting the boundaries of the image.  This worked reasonably
well but had one side effect for cursor mode as it meant the cursor
arrow could be rendered outside of the actual image dimensions.
Functionally this was fine as it was still constrained on the remote
system but visually it was not optimal.

The new approach is to use accurate boundaries for the cursor and
viewport but instead accumulate an offset value (mViewportOffset) when
the user tries to 'overpan' the image past its boundaries.  The effect
is that the cursor is still constrained to the image but we can move the
image further in either dimension to ensure they can interact with it.

Once the System UI disappears, we run a per-frame animation to reduce
the offset and reposition/render the remote image.  The effect is that
the image slides back into view quickly which is much cleaner than the
snapping effect we would have if we allowed Android to reposition the
view.

BUG= 653309 

Review-Url: https://codereview.chromium.org/2441723004
Cr-Commit-Position: refs/heads/master@{#427462}

[modify] https://crrev.com/95ccb26c84a537f696b374097f7050d472fba289/remoting/android/java/src/org/chromium/chromoting/DesktopCanvas.java

Your change meets the bar and is auto-approved for M55 (branch: 2883)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/ab6933c26cf444f61413983fb56b8700166af6a8

commit ab6933c26cf444f61413983fb56b8700166af6a8
Author: Joe Downing <joedow@google.com>
Date: Tue Oct 25 22:34:44 2016

Updating the calculations used for image overpanning

This CL simplifies the calculations used for panning the remote desktop
image out from under the System UI.  It also constrains the cursor to
the image coordinates vs. the previous method which did not.

The old approach I used was to allow overpanning the remote desktop
image by adjusting the boundaries of the image.  This worked reasonably
well but had one side effect for cursor mode as it meant the cursor
arrow could be rendered outside of the actual image dimensions.
Functionally this was fine as it was still constrained on the remote
system but visually it was not optimal.

The new approach is to use accurate boundaries for the cursor and
viewport but instead accumulate an offset value (mViewportOffset) when
the user tries to 'overpan' the image past its boundaries.  The effect
is that the cursor is still constrained to the image but we can move the
image further in either dimension to ensure they can interact with it.

Once the System UI disappears, we run a per-frame animation to reduce
the offset and reposition/render the remote image.  The effect is that
the image slides back into view quickly which is much cleaner than the
snapping effect we would have if we allowed Android to reposition the
view.

BUG= 653309 

Review-Url: https://codereview.chromium.org/2441723004
Cr-Commit-Position: refs/heads/master@{#427462}
(cherry picked from commit 95ccb26c84a537f696b374097f7050d472fba289)

Review URL: https://codereview.chromium.org/2452813002 .

Cr-Commit-Position: refs/branch-heads/2883@{#303}
Cr-Branched-From: 614d31daee2f61b0180df403a8ad43f20b9f6dd7-refs/heads/master@{#423768}

[modify] https://crrev.com/ab6933c26cf444f61413983fb56b8700166af6a8/remoting/android/java/src/org/chromium/chromoting/DesktopCanvas.java

Hey AJ, this change updates the panning behavior when the soft input method has been invoked.  We now correctly constrain the cursor to the image rect while also allowing the user to pan the image out from under the soft keyboard.

This change has been pushed into M55 as well as M56.  Can you take a look and see if you find any problems?

Thanks!

I spent some time with this and found no issues in version 55.0.2883.28

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/ab6933c26cf444f61413983fb56b8700166af6a8

commit ab6933c26cf444f61413983fb56b8700166af6a8
Author: Joe Downing <joedow@google.com>
Date: Tue Oct 25 22:34:44 2016

Updating the calculations used for image overpanning

This CL simplifies the calculations used for panning the remote desktop
image out from under the System UI.  It also constrains the cursor to
the image coordinates vs. the previous method which did not.

The old approach I used was to allow overpanning the remote desktop
image by adjusting the boundaries of the image.  This worked reasonably
well but had one side effect for cursor mode as it meant the cursor
arrow could be rendered outside of the actual image dimensions.
Functionally this was fine as it was still constrained on the remote
system but visually it was not optimal.

The new approach is to use accurate boundaries for the cursor and
viewport but instead accumulate an offset value (mViewportOffset) when
the user tries to 'overpan' the image past its boundaries.  The effect
is that the cursor is still constrained to the image but we can move the
image further in either dimension to ensure they can interact with it.

Once the System UI disappears, we run a per-frame animation to reduce
the offset and reposition/render the remote image.  The effect is that
the image slides back into view quickly which is much cleaner than the
snapping effect we would have if we allowed Android to reposition the
view.

BUG= 653309 

Review-Url: https://codereview.chromium.org/2441723004
Cr-Commit-Position: refs/heads/master@{#427462}
(cherry picked from commit 95ccb26c84a537f696b374097f7050d472fba289)

Review URL: https://codereview.chromium.org/2452813002 .

Cr-Commit-Position: refs/branch-heads/2883@{#303}
Cr-Branched-From: 614d31daee2f61b0180df403a8ad43f20b9f6dd7-refs/heads/master@{#423768}

[modify] https://crrev.com/ab6933c26cf444f61413983fb56b8700166af6a8/remoting/android/java/src/org/chromium/chromoting/DesktopCanvas.java

[Automated comment] removing mislabelled merge-merged-2840