Clarification needed regarding permissions policy integration

[bershanskiy]

The Question

As of now, the getLayoutMap() algorithm states:

If this's relevant global object's associated Document is not allowed to use the policy-controlled permission named "keyboard-map", reject the promise.

But the Privacy Mitigations section states:

[T]he API is only available from secure contexts and can only be called from the currently active top-level browsing context or granted access via policy-controlled feature.

These two statements contradict each other in the case when top-level browsing context does not have the permission.

As of October 2023, Chrome/Chromium implementation follows the Privacy Mitigations version and ignores the permission policy state in top-level frame and allows access even if keyboard-map is explicitly forbidden. Chromium issue here. As of October 2023, there is no WPT specifically for this case.

Background

Originally, getLayoutMap() was allowed only from top-level context and did not have any other restrictions. Then in #38 there was some discussion about access from sub-frames, which resulted in keyboard-map Permissions Policy.

Chronology of most important events:

• commit https://github.com/WICG/keyboard-map/commit/78cfc488458175bf0ac48c2af3737ad6f3e47647 added keyboard-map permission identifier with default allow-list 'self'
• commit https://github.com/WICG/keyboard-map/commit/9237b2630b9cf408f6a8e7c938bd1f334f78079f updated the getLayoutMap() algorithm to use the permission replacing the mention of top-level browsing context with the new permission. Please note that since the default allow list was 'self', the top-level document was still allowed access by default.
• in this commit Chromium implemented keyboard-map permission but did not remove the mention of top-level browsing context, effectively putting it in conflict with the specification. Note: the new WPT did not get a dedicated test specifically for the case of top-level context having explicit Permissions Policy keyboard-map=() and trying to access the API.
• commit https://github.com/WICG/keyboard-map/commit/bf67454b5eda610b28fa97c9aede77a7ff912bfd updated the "Privacy Mitigations" section adding wording about keyboard-map permission but not removing the mention of top-level browsing context. This was in direct contradiction with the previous commit 9237b26.
• (inconsequential) Chromium commit moved the check from Blink renderer process to the core browser process.
• (inconsequential) Chromium issue was filed specifically about this ambiguity.

Possible solutions

Remove access from top-level context when it is explicitly forbidden by Permissions Policy

• This will likely reflect the intent of the Permissions Policy system better.
• This will not change behavior for most websites since they do not specify keyboard-map permission.
• This will require a one-line change to Chromium KeyboardLockServiceImpl::GetKeyboardLayoutMap() method.
• This will require change to "Privacy Mitigations" section. PR #45 implements this change.
• This will require a WPT test.

Allow access from top-level context even in case of explicit restriction

• This will reflect the current Chromium behavior better.
• This will not change behavior for any websites at all.
• This will require no changes on Chromium side.
• This will require change to the getLayoutMap() algorithm.
• This will require a WPT test.

I can crate a Chromium CL for either solution once there is a consensus which option to take. I would favor the first solution (consistency with Permissions Policy).

Related discussion

This came up in uBlock Origin discussion.