[html-aam] Computing a role for “img”, “aside”, and “section” elements creates an infinite loop

[sideshowbarker]

Update: This is a problem for the img element as well — so I’ve updated the issue title.

---

Consider these two things:

A. Computing a role for an aside or section element. The conditions at https://w3c.github.io/html-aam/#el-aside and https://w3c.github.io/html-aam/#el-section for computing a role for an aside element or section element in turn require computing an accessible name for the aside element or section element.

B. Computing an accessible name for an element. Multiple steps in the algorithm at https://w3c.github.io/accname/#computation-steps for computing an accessible name for an element in turn require computing a role for that element.

That is: In order to implement A. Computing a role for an aside or section element in my code, I need to have that code call my implementation of B. Computing an accessible name for an element. But in order to implement B. Computing an accessible name for an element in my code, I need to have that code call my implementation of A. Computing a role for an aside or section element.

In other words, my code gets into an infinite loop if I implement the requirements as specified.

See also Siteimprove/alfa#298

Link to documentation: https://w3c.github.io/html-aam/#el-aside

Does the issue exists in the editors draft is the most recent draft of the specification? Yes

[smockle]

B.…computing an accessible name for an element in turn require computing a role for that element.

aside has either a complementary or generic role. Since generic can’t be named, when checking whether aside has an accessible name, its role can be assumed to be complementary, right? By assuming that, actual role computation would be unnecessary, avoiding the infinite loop.

[scottaohara]

would this make more sense if the spec indicated the default roles to be complementary / region - and THEN have a condition to change those roles based on ancestral checks (for aside) or if the elements are named (for both elements) - rather than separating these out so the elements each had two tables?

[annevk]

I think you should resolve the loop in the algorithms somehow. E.g., give "computing an accessible name" a boolean that indicates it's invoked for role computing purposes in which case it uses fallback roles for certain elements might do the trick?

[sideshowbarker]

w3c/aria#2404 describes a similar problem I found in another part of the ARIA spec. To resolve it, I ended up needing to do as AnneVK suggested in #570 (comment) — that is, I needed to add a boolean parameter to my implementation of the “computing an accessible name” algorithm, so that callers can specify they don’t want role computation (re)performed.