[whatwg] tabindexscope

[Ian Hickson]

On Mon, 30 Jan 2012, Tab Atkins Jr. wrote:
> On Mon, Jan 30, 2012 at 1:54 PM, Ian Hickson <ian at hixie.ch> wrote:
> > On Tue, 8 Nov 2011, Ojan Vafai wrote:
> >> We keep running into the use case where the physical position matters
> >> for the tab order. The problem with just setting tabIndex (or CSS3
> >> tab-index) is that it takes the thing out of the natural order.
> >>
> >> This problem comes up in a lot of places (e.g. absolute positioning).
> >> It's recently come up for CSS flexboxes, e.g. if you set flex-order or a
> >> reverse flow, then the tabindex still being in document order is often
> >> not what the author wants
> >> (https://bugs.webkit.org/show_bug.cgi?id=62664).
> >>
> >> <button tabindex=0>A</button>
> >> <div tabindex=2 tabindexscope>
> >> <button tabindex=2>C</button>
> >> <button tabindex=1>B</button>
> >> </div>
> >> <button tabindex=1>D</button>
> >>
> >> The order for the tabbing would be A-D-B-C.
> >
> > The spec says that the order when you omit tabindex (or set it to 0)
> > should follow platform conventions. If the platform convention is to make
> > the tab order follow the visual position, then that's what the browser
> > should do.
> >
> > Surely that would be better than having authors manage local regions for
> > tabindex, especially since the positioning depends on the CSS level, not
> > the HTML level, and thus trying to manage the tabindex in the HTML would
> > be a layering violation anyway.
> 
> If you are attempting to match the tab order to the position of an
> element, you are correct.  In this situation, the tab order of the
> group itself should be controlled by the 'nav-index' property
> alongside the positioning code.
> 
> However, *within* a group of controls, the relative order can want to
> be scoped without reference to CSS.  This can happen because the group
> is being positioned with CSS (and thus the appropriate tab-index is
> unpredictable), because the group may be generated into multiple pages
> with different tab-index'd items elsewhere in the page, or just
> because the dev would like to write their tab-indexes without having
> to renumber everything every time they move the HTML around in the
> page.
> 
> Scoping a tab-index is thus a property that can appropriately belong
> to the HTML level, just as much as tab-index itself does.

Can you give some examples of real-world pages where the tabindex 
attribute has been used (with difficulty due to the lack of scoping), 
where nav-index is not the right solution, and where the UA following 
platform conventions for tab order doesn't or wouldn't end up in a good 
UI, that show that this feature would be useful? I'm having trouble 
picturing it, and the frequent references above to positioning and other 
CSS layout features is confusing me.

-- 
Ian Hickson               U+1047E                )\._.,--....,'``.    fL
http://ln.hixie.ch/       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'