Recent additions to ChromeClient added default implementations. From a porter point of view this is rather bad as old features might break and no runtime warning is printed and searching for notImplemented will not show anything. The approach to treat the *Client as an interface and only have pure virtual is a lot better in this regard. Change ChromeClient to be a interface again.
Created attachment 30411 [details] Make recent additions pure virtual again Make setCursor, scrollRectIntoView, requestGeolocationPermissionForFrame pure virtual.
Comment on attachment 30411 [details] Make recent additions pure virtual again > - virtual void scrollRectIntoView(const IntRect&, const ScrollView*) const {} // Platforms other than Mac can implement this if it ever becomes necessary for them to do so. > + virtual void scrollRectIntoView(const IntRect&, const ScrollView*) const = 0; // Platforms other than Mac can implement this if it ever becomes necessary for them to do so. I don't think the comment makes sense any more if this is a pure virtual. This change seems OK. This is the reason I used pure virtuals in the first place, but I'm not entirely sure either way whether this is a good thing or not. r=me
Changed the comment to mention the implementation for mac and landed in r43918. With my "porters" hat on I like the pure virtual functions, they move everything relevant to the various client implementation that can be found in the kit...