there's no criteria :) Firefox just makes them accessible. Perhaps, we should just ignore them until they have explicit semantics like @role attribute.

Span is not mapped in IA2 or ATK, so it would seem consistent to do the same for div. Is there benefit to giving them an object mode in the tree?

Joanmarie, any opinions?

I also note that FF is still using the BSTR hack and setting tag name for MSAA accrole. Chrome does it too, but with fewer elements. I seem to remember some talk about discontinuing that, at least by some Mozilla folks (might just have been you and David B). Are some AT using that?

(In reply to Jason Kiss from comment #2)
> Span is not mapped in IA2 or ATK,

That is the case when there is nothing other than text content and possibly text attributes associated with that span. Give a span an onclick handler, a title, or something similar and then it should be mapped.

But like you said, for the majority of instances, it is not mapped. And the content it contains gets folded into the parent object which is usually a render block.

> so it would seem consistent to do the same for div.

Depends on the div. :)

1. <div onclick="foo();">click me</div>
2. <div>bunch of text like a paragraph but it's a div</div>
3. <div id="useless"><div id="useful">bunch of text</div></div> 

For 1, clearly it needs to be exposed.

For 2, I think it should be exposed because it's a distinct block of text.

For 3, you can probably guess what I think. ;) But the degree to which the hierarchy can safely be flattened is an ongoing debate, one that's lasted for years and shows no signs of concluding at a point of consensus. Ask Jamie or Alex. ;) I no longer have the energy to push for flattening* and welcome our DOMy embedded object character overlords.

* Still have the energy to be snarky. ;)

(In reply to Jason Kiss from comment #2)
> Span is not mapped in IA2 or ATK, so it would seem consistent to do the same
> for div. Is there benefit to giving them an object mode in the tree?

technically divs can be used to create visual blocks that has to be accessible, but you can argue that the author has to take care about that explicitly. Also accessible div makes sense if it's a text container.

> I also note that FF is still using the BSTR hack and setting tag name for
> MSAA accrole. Chrome does it too, but with fewer elements. I seem to
> remember some talk about discontinuing that, at least by some Mozilla folks
> (might just have been you and David B). Are some AT using that?

iirc there's couple instances that AT relies on, major number of cases doesn't need that

I think I can prototype these in Firefox and see how it goes. If it doesn't have any negative consequences on the web then we can make spec more restrictive.

HTML AAM is no longer an ARIA Deliverable. If you wish to pursue this issue please file at https://github.com/w3c/html-aam/issues