Clarify the implications and limitations of aria-owns

[WilcoFiers]

This follows a conversation with @smhigley here: #1454 (comment)

That PR is doing an excellent job of clarifying the intended parent/child relationship between certain roles, and it introduces a definition for owned children. This raises the question though of who owns an element. Take this example:

<ul id="list1"> <li id="item">Hello</li> </ul>
<ul id="list2" aria-owns="item"></ul>
<ul id="list3" aria-owns="item"></ul>

The way the spec is currently phrased suggests an element can be owned by more than one element. I.e. there is no exception saying that an element referenced with aria-owned is no longer owned by its ancestors too. There's also nothing suggesting that aria-owns can't be used on the same element more than once. Common sense suggests the same element can't be in the tree more than once. It would be good to update the spec to be explicit about this.

There are also some other caveats to consider, like elements referencing themselves, or elements causing recursive owning that should probably explicitly be prohibited:

<ul id="list" aria-owns="item"> <li> Item 1</li> </ul>
<li id="item" aria-owns="list"> Nested item 2 </li>

[jnurthen]

The way the spec is currently phrased suggests an element can be owned by more than one element. I.e. there is no exception saying that an element referenced with aria-owned is no longer owned by its ancestors too.

I believe this is implied but not explicitly stated. It should be.

There's also nothing suggesting that aria-owns can't be used on the same element more than once. Common sense suggests the same element can't be in the tree more than once. It would be good to update the spec to be explicit about this.

"Authors MUST ensure that an element's ID is not specified in more than one other element's aria-owns attribute at any time. In other words, an element can have only one explicit owner."

is more needed?

There are also some other caveats to consider, like elements referencing themselves, or elements causing recursive owning that should probably explicitly be prohibited:

"Authors MUST NOT create circular references with aria-owns." is already in the spec for aria-owns. Is this not enough?

[spectranaut]

Aaron's notes about the limitations of aria-owns: https://docs.google.com/document/d/1jhztcbvUWS4_QGtKmWMxVBLnMkl5ZAAqRCVTcLfdl4o/edit#heading=h.4qp85atck7yr

[WilcoFiers]

@jnurthen I think the most important thing here is for ARIA to be explicit about an element only having one owning parent. But beyond that I think it would be quite helpful if ARIA could state what user agents should do with invalid ARIA. I don't think it's enough for a standard to say content authors shouldn't make mistakes. They're going to. It helps to know what browsers should do when they made such mistakes. ARIA has some presidence for this in the forms of presentational role conflict.

HTML 5 also defined how to handle malformed HTML. Doing so made a big different to accessibility. I think the same applies here.