| Chapter I: New features |
| Composite background-color animation |  |
 |
 |
 |
 |
 |
 |
xidachen@ launched finch study for this feature and this doc summarizes its result. Here are some high-level points: | We are seeing more animations being composited. | Unfortunately we don’t see improvement in the “PercentDroppedFrame”, which reflects the smoothness of the animations overall. One possible reason is that background-color animation is mostly applied on small buttons and it is not the dominant factor of dropped animation frames. | We are seeing memory regression and this is tracked by crbug.com/1210221. | Moreover, this feature is now experimental, which means one can enable it by enabling “Experimental Web Platform features” in chrome://flags. | Capability Delegation Spec |  |
mustaq@ finished the draft spec (link) and it is ready for review. | Android 12 Overscroll |  |
flackr@ has finished the initial version of working overscroll. | It re-uses elastic overscroll physics from Mac overscroll. | Instead of translating, it applies overscroll stretch on the transform node. | There are lots of follow-on work including: | crbug.com/1213217 | crbug.com/1213248 | crbug.com/1213252 | Enable only on Android-S |
| Chapter II: Stability fixes |
| Fix privacy leak of visited links |  |
kevers@ fixed a privacy leak issue of visited links. A repro case is shown above. The fix to the issue is: | Extract the visited style even if link is unvisited during style resolution. | Generate a transition if either the visited or unvisited style changed. | animation.effect.getKeyframes reports unvisited style regardless of whether visited. | Visually render with the correct style. This is possible since the interpolation code retains a pair of colors to interpolate corresponding to the visited and unvisited style. | Fix a crash in Mac IME fix |  |
flackr@ fixed a crash by performing a null check. | Removed one fake user activation |  |
mustaq@ fixed a fake activation issue as a suspect for a navigation problem. | UMA investigation revealed <0.005% page loads could be affected. | Tentatively removed the fake user activation notification from JS play() and everything seems good as before! |
| Chapter III: Interop fixes | Overscroll-behavior |  |
xidachen@ kept working on the bug, where chrome propagates overscroll-behavior from the <body> element, while by spec it should be from the <html> element. | A detailed doc describing the problem. | Break the feature counter into two cases. (CL) | Querying UKM table shows that google.com is the major site that have this problem. |
| Chapter IV: à la carte | Bisecting on Android |  |
skobes@ found that it is quite annoying to bisect on Android, but it is possible, and also that the tool could be improved. | go/chrome-android-bisect contains all the details on how to bisect on Android. | skobes@ used that for two recent top-controls regressions | crbug.com/1167400 | crbug.com/1207888 | AtomicString for PointerEvent type |   |
flackr@ replaced char\* with AtomicString for PointerEvent type. Why? | The returned string is passed to PointerEventInit::setPointerType implicitly cast to a WTF::string for the rest of its existence. What’s the problem? | First, we take the strlen of the string, which is O(n) operation. | Then, we allocate storage space and copy the string constant, which is O(n) time + space. | If we instead construct a String from another String, then it just adds a ref to the StringImpl, and that’s O(1). Refer to the screenshot on the left hand side for details. | Now the String is constructed, are we done? Well, not yet, we have to compare two strings. | Comparing two arbitrary Strings is O(n), see EqualStringView implementation. | But, comparing two AtomicStrings is O(1)! Refer to the screenshot on the right hand side for details. |
| Chapter V: Bug Updates |   |
Our team kept on top of bugs. We closed almost the same amount of P1 bugs vs opened P1 bugs. |