When going fullscreen on a Mac with a 1680x1050 display, window.innerHeight reports 974, and not 1050. On Win7, on a 1920x1200 display, innerHeight is correctly 1200. There's a simple testcase for this at http://people.mozilla.com/~vladimir/misc/fs.html . Note that bug 712225 might be relevant -- that also has a similar testcase at http://diogogmt.github.com/mozilla-central/windowInnerOuterHeight.html (move the mouse to get the numbers to update). We should fix this and figure out what it affects, as it can have a pretty huge impact on gaming viability (since fullscreen elements need to be resized to get 1:1 pixel ratio).

What you're seeing is the "mozfullscreenchange" event firing before the transition to fullscreen is complete. Specifically before the window manager believes the chrome menubar is hidden. If you check window.inner* after the "mozfullscreenchange" event (as this attached testcase does) you'll see we return the correct value at some time after "mozfullscreenchange". I was working in bug 756334 to address delay dispatching "mozfullscreenchange" until the transition was complete, but got bogged down with mochitest timeouts on tinderbox. It may be a month or two before I get back to working on that...

Is this a regression? Why should we track for FF19 specifically?

This is not a regression. This issue has existed for as long as we've had a fullscreen API.