STR:

1. Visit t.m.o. This shows GC_MS for me.
2. Open "Advanced Settings".
3. Scroll down a little.
4. Check "Cumulative Mode".

Expected:

The checkbox is checked. The graph is updated.
The mouse cursor works as before. When you click on something on the page, that is the thing that receives the event.

Actual:

The checkbox is checked. The graph is updated.
The mouse cursor is offset downwards by in excess of 100px.

The place that the cursor appears is correct (it does not jump around when moving to other windows), but the location reported to the page is not. The page shows hover tips for the graph when the cursor is still within the "Advanced Settings" box. The cursor changes to 'pointer', activating the :hover on the "Advanced Settings" header, somewhere in the middle of the box.

Attempting to clear the checkbox has no effect as the click is being reported as hitting a blank space 100px or more above the checkbox. It is possible to activate elements on the page that are off the screen.

Leaving and returning to the window has no effect. Any amount of scrolling corrects the problem.

This does not happen if the top of the page is in the viewport when the checkbox is selected, or only a very small amount of scrolling has occurred. The effect is smaller if only a small amount of scrolling has occurred, but a large amount of scrolling does not seem to make the problem any worse. Maybe this is limited by the tile size used by the compositor.

Confirmed in a clean profile with webrender on, APZ fully enabled, and OMTP on. NVIDIA acceleration.

The severity field is not set for this bug.
:mattwoodrow, could you have a look please?

For more information, please visit auto_nag documentation.

I'm going to try moving this to APZ for now, since it seems like the content process might be receiving coordinates that have been adjusted incorrectly.

The symptoms sound similar to bug 1657627, could be the same underlying issue.

Yeah, I confirmed it's the same problem with the same regression range.

I'm going to use bug 1657627 as the canonical one to track this problem, but will verify this bug as well once the fix is landed.

Verified this is fixed in the latest nightly.