Specifically, I'm expecting some calls to set scrolling behaviours on LayerTreeHost for such frames from content and/or blink, in a similar way to when there is an event handler. (Do we just say there is an event handler or?)

Alternatively, popups disable compositor thread scrolling also through a few mechanisms and perhaps that should be unified with non-main-frame local roots (RenderWidgets)?

Ah yes, I remember this one. No one was happy about it but IIRC this was added as a quick hack to get OOPIFs working - I believe this was because we needed a different scroll status so that we get the correct scroll bubbling behavior between OOPIFs (i.e. the scroll wasn't handled but should bubble to the parent, rather than being dropped). All the other uses are DCHECKs added to tease out OOPIF bugs.

It should be possible for an OOPIF and non OOPIF renderer should be returning the same scroll status if we reached the end of the scroll chain without scrolling - the parent/browser should decide whether to bubble to the next renderer. I think this was always the desire but required some thought and we were in a rush to ship OOPIF.

+wjmaclean@, is there someone on the OOPIF side who can look into cleaning this up? (Also, returning UNKNOWN as a status to bubble has always bugged me)

I'm happy to remove it, though this is blocked on resolving the PSF != 1 in OOPIFs bug, since it's the best (only?) diagnostic tool I have for it.