Boris Zbarsky wrote:
> Right. Any time implementations disagree, the most likely conclusion is
> that the point on which they don't agree is not used much, if at all, by
> websites. Therefore the exact behavior the spec requires on that point
> doesn't affect the "compatible with the web" criterion.
Agreed. In which case it makes sense to either leave the specifics
undefined, or to standardize the behavior that makes the most sense.