+rhalavati@

As you probably know, the side effect you're describing is exactly the reason for the current behavior. Changing the default value of a content setting is relatively rare (since that's deep in the settings), and changing a content setting exception before you visit a site would be even rarer.

So while not a deterministic indicator of Incognito, this would become a high-probability heuristic. Similar to how Facebook infers Incognito from the fact that your cookie jar is surprisingly free of their tracking cookies. And with that, it could end up annoying users who really did block notifications.

I'd prefer that we instead explore ways how we can support notifications in Incognito, addressing both the developer point of view and the privacy requirement that web-facing behavior of Incognito should be indistinguishable.

Hi,

Sorry for the delay, I dug up the previous discussions on this issue in:
 crbug.com/401439 ,  crbug.com/479679 , crbug.com/542081, and summary in https://docs.google.com/spreadsheets/d/1BI3Lolm1-8V700_aXgePhAjQVAgTPYpUH2ewc59XKRM

I did not see any argument against implementing Push notifications API in incognito, except for it's not being useful, useless prompts to the user, and clogging up web app's server ( crbug.com/401439#c2 ).

For not using the default DENIED case, the main argument has been what Martin mentioned before, statistical correlation between immediate deny and being in incognito.

So IMHO, a full implementation is the best choice from user experience point of view and also can be useful for developers or testers who need a fresh profile. 

But if we don’t agree or we don’t see implementation of it worth the effort, we should consider that privacy is user’s main concern when s/he is in incognito and sacrificing some level user experience for the sake of better privacy can be the better choice. So I also prefer simulated denied to defaulted denied as it has less chance of revealing the incognito mode.

Okay. Thanks for the input - let's close this. It's still not feasible to implement Incognito push today due to server-side costs, but those are being worked on with an ETA of later this year.