- IMPROVED: on mobile devices where the address bar shows/hides and sometimes causes a jump after scrolling stops, particularly if there are multiple pins, ScrollTrigger now calculates positions based on 100vh instead of window.innerHeight because the former doesn't change whereas the latter does. This essentially avoids the "jump" in most cases (unless the elements on your page actually resize).

- IMPROVED: consolidated "scroll" event listeners in ScrollTrigger so fewer are added to the main document. Functionally it makes no difference, but there is at least one synthetic test online that would [incorrectly] complain.

- FIXED: regression in PixiPlugin caused an error related to filters. See https://github.com/greensock/GSAP/issues/541

- FIXED: regression in 3.12.0 caused ScrollTriggers that were applied to timelines inside a gsap.context() not to get reverted properly when revert() was called on the context. See https://greensock.com/forums/topic/37278-updating-to-312-3121-breaks-pinning/

- FIXED: if you insert a tween/callback into a timeline at a position EARLIER than where the playhead currently is, it may not render immediately, or if it's a callback it may fire on the next render when it actually shouldn't. See https://greensock.com/forums/topic/37218-timeline-trigger-all-callbacks-with-specific-timing-when-play/

- FIXED: regression in 3.12.0 caused an error in Draggable if you enable inertia and then disable() the Draggable instance. See https://greensock.com/forums/topic/37397-draggable-error-in-version-12-when-using-disable/

- FIXED: regression in 3.11.0 could cause a repeating tween/timeline to render at the wrong iteration if it was reversed and then a totalTime() was set that was beyond the first iteration.

- FIXED: if you revert() a gsap.context() that has a reversed animation that'd cause its startTime to be lower than -1 and it had .fromTo() tweens, the "from" revert happened out of order which might have left the starting values incorrect.

- FIXED: regression in 3.12.0 could cause a ScrollTrigger's snapping to act incorrectly in a very specific scenario, like if mid-scroll the start/end values change.