By default, the browser only sends the brand, major/significant version, and whether or not the browser has signalled it wants a "mobile" experience. Other than that, the browser only sends hints that the site has signalled it wants through the Accept-CH response header.

If, for example, the browser has never visited the site before, then the browser has never received an Accept-CH response header from that site and therefore doesn't know that the site wants any extra headers. This also happens when information about the site is clear, which also clears the client hint preferences

It could also be that the site has changed the hints it wants to receive since you last visited, in which case you might not have sent all of the headers it actually wants.

In both cases, it's not that the browser is blocking the headers, but that the preferences stored in the browser are out of sync with the preferences of the site.

On Thu, Jan 28, 2021 at 10:53 PM Ryosuke Niwa <rniwa@webkit.org> wrote:
I'm still confused here. In what scenario would a browser decide to not send client hints but later decide it's okay to send them?

On Thu, Jan 28, 2021 at 7:13 PM Aaron Tagliaboschi <aarontag@chromium.org> wrote:
The Critical-CH header can trigger a request re-try. It's for situations where the browser could be unaware of the site's CH preferences (like the first navigation request to a site before the browser has received and stored CH preferences) or if a site has changed those references, and the site would rather drop the request and retry over getting a potentially "incomplete" request

This would *not* override potential mitigations or reductions in fingerprinting surfaces imposed by the browser. Any headers that would be blocked would still be silently dropped.

(cc davidben, mjs who I forgot to CC the first time)

On Thu, Jan 28, 2021 at 9:35 PM Ryosuke Niwa <rniwa@webkit.org> wrote:
What's the point of specifying Critical-CH as opposed to relying on CH provided by the browser?

Is the idea that some browsers may decide to hide some client hints to reduce the fingerprinting surface?
If so, then this new header seems to just defeat that because a website can specify all the client hints as critical.

- R. Niwa

On Wed, Jan 27, 2021 at 4:40 AM Aaron Tagliaboschi via webkit-dev <webkit-dev@lists.webkit.org> wrote:

The Client Hint Reliability proposal is a set of features aimed at making Client Hints 
<https://tools.ietf.org/html/draft-ietf-httpbis-client-hints-15> more reliably available and mitigating 
mis-matches between a site's preferences and the preferences stored in the browser. The idea 
behind the Critical-CH response header is to signal to browsers that there are hints the server 
would rather pay a round trip than not have not the first request. The basic algorithm is as follows:

If, after receiving a request with Critical-CH and Accept-CH headers, there is a hint indicated in 
the Critical-CH header that the browser did not send but would not block sending, the browser 
should store the new CH preferences, drop the request, and start a new one with the new 
headers included.

Aaron Tagliaboschi | Software Engineer, Chrome Trust & Safety
_______________________________________________
webkit-dev mailing list
webkit-dev@lists.webkit.org
https://lists.webkit.org/mailman/listinfo/webkit-dev