Tabs infinitely rerenders when selectedKey is null

[abeljohn]

Provide a general summary of the issue here

When setting the selectedKey on RAC Tabs to null, Error: Maximum update depth exceeded. is thrown repeatedly and the browser freezes up due to the infinite rerender.

🤔 Expected Behavior?

Tabs should render with no Tab selected when selectedKey is null, since null is allowed by the SingleSelection types

😯 Current Behavior

Tabs infinitely rerenders when selectedKey is null and throws errors repeatedly.

💁 Possible Solution

Seems to be a regression in RAC 1.14. I bisected RAC nightlies and found it was introduced between 3.0.0-nightly-0836d3a4e-251017 and 3.0.0-nightly-02433d1f1-251018, so most likely #9041 and the resulting extra rerender. However, this seems like a logic issue in useTabListState so I think it can be fixed there

A workaround is to set the selectedKey to a nonexistent Key instead of null in order to control the "no tab selected" state

🔦 Context

There is a case where Tabs state is controlled and initially nothing is selected, so it seems reasonable to pass null as the selectedKey

🖥️ Steps to Reproduce

https://codesandbox.io/p/sandbox/peaceful-jackson-lxxwyr (open the browser console)

Version

RAC 1.14.0

What browsers are you seeing the problem on?

Chrome

If other, please specify.

No response

What operating system are you using?

macOS 15

🧢 Your Company/Team

No response

🕷 Tracking Issue

No response

[snowystinger]

Thanks for the issue and PR. In Tabs you must always have one selected tab. This is an issue in our types.
https://www.w3.org/WAI/ARIA/apg/patterns/tabs/

When a tabbed interface is initialized, one tab panel is displayed and its associated tab is styled to indicate that it is active. When the user activates one of the other tab elements, the previously displayed tab panel is hidden, the tab panel associated with the activated tab becomes visible, and the tab is considered "active".

What are you trying to build? Maybe we can recommend a different pattern that would be more appropriate.

Would you like to update the PR to omit the 'null' from our types?

[abeljohn]

@snowystinger thanks for the reference to the spec. I don't have a super strong use case here, it's just something that was supported previously in my project because it was supported by RAC. I'm all for a11y compliance, so I think updating the types to enforce this (and avoid the state where the page crashes) is a good idea. I'll update the PR accordingly