hit regions and a11y geometrics

Hi. We've got a discussion at Mozilla bug [1] about drawFocusIfNeeded spec
and implementation. I'll try to outline the outspoken concerns there and
sketch up the alternative approach.

A first thesis I would make is the web author can do CSS to define geometry
of DOM elements underneath the canvas and in most cases it should work
well. Having said that I admit that it still may be useful to have special
API to define the boundaries since
1) it may be just handy (if you drawn an element then reuse its path for
a11y needs)
2) geometry of drawn thing may be different from geometry of associated DOM
element (like fancy triangle checkbox)

As I understand a current approach adopted/implemented by Chrome is to make
drawFocusIfNeeded to define accessible boundaries for any element not
depending on whether the focus is drawn or whether the element is focused
or not.

I see an advantage of the approach in its simplicity like the author can
"get accessibility for free" but imo the same time this advantage may be
considered rather confusing:

1) the method name doesn't make it evident that accessible object
boundaries are changed (especially when it's called on the element having
no focus), so the author theoretically can make wrong assumption about the
method behavior
2) A focus ring is not really the same as element boundaries, for example,
focus ring created by CSS outline style is drawn outside the element
boundaries. Having said that I agree it must be close and may provide a
good idea about object boundaries.

On the other hand the spec provides hit regions concept that can be used to
define accessible boundaries. For the author it means he have to make one
extra call (drawFocusIfNeeded + addHitRegion).

Besides boundaries definition the hit regions are useful to connect the
canvas representation with its DOM element. In other words, if canvas
representation gets clicked then DOM element gets clicked too so the
assistive technology gets working a11y for free. The converse is true. If
the assistive technology invokes an action on DOM element then its canvas
representation gets clicked too and all visual stuff just works.

For me it seems like we have a piece of technology that provides a bunch of
good accessibility features including the setting of accessible boundaries
and it's quite tempting to prototype it. What concerns to ability of
drawFocusIfNeeded to define boundaries then having hit regions this ability
rather complicates the things.

Thank you.
Alexander.

[1] https://bugzilla.mozilla.org/show_bug.cgi?id=958241