<body> <input tabindex=30 value=30> <input tabindex=20 type=date> <input tabIndex=10 value=10 autofocus> </body> Steps to reproduce: 1. Open the document above on Google Chrome 24+. 2. Press TAB 3 times ==> The last sub-field in the date input is focused. 3. Press TAB Expected: The input with tabindex=30 is focused. Actual: The 'Back' navigation button is focused. The shadow host of the date input is not focusable, and it's shadow sub-fields are focusable. We have no ways to control focus sequence in this case though the shadow host has tabindex. Possible solutions: A) Introduce special handling to respect tabindex on non-focusable shadow host B) Make the shadow host <input> focusable In this case, we'd like to focus on the first sub-field when the shadow host is focused, and Shift+TAB on the first sub-field should skip focusing on the shadow host. We need another hack to detect focus direction. C) Stop using shadow DOM focus navigation. Make the shadow host <input> focusable, and make sub-fields non-focusable. We can emulate the current behavior by detecting focus direction.
Just to make sure I understand, let me try to say it back to you in a slightly different way: We need a way to indicate that some shadow hosts must never participate in sequential focus navigation, delegating their focusing completely to the tree they host. Right?
(In reply to comment #1) > We need a way to indicate that some shadow hosts must never participate in sequential focus navigation, delegating their focusing completely to the tree they host. > > Right? I think so. I feel the current rule for focusable-shadow-nodes-in-non-focusable-shadow-host has a defect because component users have no ways to specify tabindex to a component. We might want to introduce a new attribute for shadow hosts like shadow-tree-tab-index=20. Anyway, we need a hack for input[type=date] because it should work with normal tabindex attribute.
(In reply to comment #0) > <body> > <input tabindex=30 value=30> > <input tabindex=20 type=date> > <input tabIndex=10 value=10 autofocus> > </body> > > Steps to reproduce: > 1. Open the document above on Google Chrome 24+. > 2. Press TAB 3 times > ==> The last sub-field in the date input is focused. > 3. Press TAB > > Expected: > The input with tabindex=30 is focused. > > Actual: > The 'Back' navigation button is focused. > > > The shadow host of the date input is not focusable, and it's shadow sub-fields are focusable. We have no ways to control focus sequence in this case though the shadow host has tabindex. > > Possible solutions: > A) Introduce special handling to respect tabindex on non-focusable shadow host > B) Make the shadow host <input> focusable > In this case, we'd like to focus on the first sub-field when the shadow host is focused, and Shift+TAB on the first sub-field should skip focusing on the shadow host. We need another hack to detect focus direction. B) was what I had implemented before. I called it 'focus delegation'. But I've removed that to prefer a general rule defined in the spec. I think we need a hack to keep a compatibility for built-in elements, such as input elements, which has a shadow tree unless we can have a good spec to cover this case. > C) Stop using shadow DOM focus navigation. > Make the shadow host <input> focusable, and make sub-fields non-focusable. > We can emulate the current behavior by detecting focus direction.
I discussed this with some people offline, and concluded we should take B. A is not good because we need to add a hack to FocusController only for INPUT_MULTIPLE_FIELDS_UI. C is not good because we'd like to expose aria-* attributes on sub-fields. If sub-fields were not focusable, WebCore a11y would not expose them. To implement B in HTMLInputElement, I'd like to add FocusDirection argument to Node::setFocus.
Created attachment 185981 [details] WIP
Comment on attachment 185981 [details] WIP Attachment 185981 [details] did not pass mac-wk2-ews (mac-wk2): Output: http://queues.webkit.org/results/16309541
Comment on attachment 185981 [details] WIP Attachment 185981 [details] did not pass mac-ews (mac): Output: http://queues.webkit.org/results/16298545
Created attachment 186315 [details] Patch
Comment on attachment 186315 [details] Patch Attachment 186315 [details] did not pass win-ews (win): Output: http://queues.webkit.org/results/16377009
Comment on attachment 186315 [details] Patch (In reply to comment #9) > (From update of attachment 186315 [details]) > Attachment 186315 [details] did not pass win-ews (win): > Output: http://queues.webkit.org/results/16377009 false alarm.
Comment on attachment 186315 [details] Patch r=me
Comment on attachment 186315 [details] Patch Clearing flags on attachment: 186315 Committed r141835: <http://trac.webkit.org/changeset/141835>
All reviewed patches have been landed. Closing bug.