56.0.2924.79 - good
57.0.2955.0 - good
57.0.2971.0 - good
57.0.2972.0 - good
57.0.2973.0 - bad
57.0.2975.0 - bad
57.0.2979.3 - bad
57.0.2987.9 - bad

https://chromium.googlesource.com/chromium/src/+log/57.0.2972.0..57.0.2973.0?pretty=fuller

Re-adding Needs-Bisect; can we get an exact bisect please?

I repro in Chrome, too. I don't know what component that makes this. Maaaybe compositing?

In the regression range is the composited border radius patch which would tend to clip content (although why this content is unclear). 

Yeah, did bisect to https://codereview.chromium.org/2588853002.

Note that it also makes scrolling jank significantly (which was what I was bisecting for). Looks like rastering is taking a lot of time on the gpu thread after that change. More obvious on a slower device like nexus 5x than on a pixel.

Could someone check if this is fixed in the latest Canary?

If not fixed, does it repro in mobile emulation on desktop? I'm away from my usual dev setup so cannot easily verify these things.

Clipping appears to be fixed for me in 58.0.2999.0; the jankiness still seems to exist though.

What CL did you think fixed the clipping?

https://chromium.googlesource.com/chromium/src/+/2d6bace7c6b2ea719eb8dac4f0fca2e0423c18a1

The "fix" avoided the bad clip in cases where it is not needed, that is most cases. It does not fix the underlying issue for cases where the clip is needed, so we should leave this open so I can try to fix the underlying issue.

And jankiness might be unrelated. I can try to verify that when I get back to my home desk.

I can't reproduce the masking issue even with my optimization patch disabled and reverting the page back to having border radius on the hover div (that had been removed). Scrolling performance on my Nexus 5x is fine and also fine on 57.0.2987.38 on my test Nexus 4.

Marking WontFix as it is no longer reproducible.