Or to the users of WebLayerImplFixedBounds, what would that look like?

I think most of the users are actually WebLayer users (such as GraphicsLayer). So, propagating this there might look awkward... I think it would be the difference bewteen

if (webLayer->isFixedBounds())
  webLayer->invalidateRect(rect);
else
  webLayer->invalidate();

anywhere we call invalidate (2 callsites)

instead of having

Layer::SetNeedsDisplayRect(gfx::Rect rect) {
  if (fixed_bounds_)
    rect = gfx::Rect(bounds()); // (since SetNeedsDisplay just calls
                                //  SetNeedsDisplayRect(gfx::Rect(bounds()))
  ...
}
    
  
I would prefer to try to move this into Layer. Do you have reasons for moving it the other way?

I think in practice SetNeedsDisplay doesn't get called on fixed bounds layers.  The real difference is that it doesn't change bounds, which normally creates implicit invalidations elsewhere.

I do think we should remove this class at this point.  It was a convenient way to do things in the past, but I think it's going to get in the way in the future.

(I was just using SetNeedsDisplay as an example, similar transformations would apply to SetTransform and SetBounds)

The latest idea here (credit: enne) is to have a raster_bounds as well as regular bounds on a layer. Then from layer to layer_impl, we can just push properties as usual. However, when we get to layer_impl (specifically picture_layer_impl), we can use raster bounds when creating the tilings and then scale the resulting tiles to fit into bounds when we append quads.

enne@ please correct me if I'm wrong. 

I still plan on taking a look at this, but not in the immediate future.