Replace device-info permission by a browsing context boolean flag

[youennf]

No description provided.

[alvestrand]

I disagree with this approach.
I think it has no advantages over the permission-based approach, and requires special-casing situations where we want to have device-enumeration permission separate from device-access permission, rather than just saying "and you can grant this permission in that case too".

Please file a bug describing the problem you want to solve.

[youennf]

The bug is #612.
It is not great that because I give access to a camera once to take a picture of me, the website will be able to get all my device information for my future visits.

Also, we agreed that the usual good approach is to do getUserMedia-then-enumerateDevices.
This PR is consistent with that approach.

Additionally, I am not sure the device info permission concept is good.
Other permissions are things like "can I use a camera", which is easy to describe to a user.
This specific permission is hard to explain to a user and AFAIK, is not exposed by any UI, contrary to camera/mic access.

[jan-ivar]

This PR removes an arguably redundant permission, which I approve of.

But it still says user agents may revoke this ability (permission) whenever they want. This could be:

1. Never
2. During actual camera or mic capture, sabotaging post-gum in-content selection

I think we need to define this in more detail to protect web interop.

Original intent: when users revoke both camera and microphone permission, they revoke device-info as well (not before). I.e.

device_info = camera_permission || microphone_permission

This is true in all browsers I tested, so it must have been the original intent. Other parts like groupId support this notion. I vote we tighten the spec to conform with implementations here.

This was also meant to be extensible, which is where I share @alvestrand's concern. E.g.

device_info = camera_permission || microphone_permission || speaker_permission

If we end up not needing this because of w3c/mediacapture-output#86, great! But it would still be nice if this spec didn't forbid extension, I think—Forbidding user agents from doing whatever they want is not the same as forbidding other specs from extending things.

Is there a way to write it like this?

[jan-ivar]

I understand permissions may be per-device, so if this needs to be per-device that would WFM.

[alvestrand]

Needs a PR to remove device-info from feature-policy. @youennf has promised to do this.