Rewrite :visited rules to match reality, based on dbaron's spec.

[tabatkins]

Per #2037, adapts dbaron's spec into more generic language, and adds several examples and some surrounding verbiage.

[FremyCompany]

Approved with suggestion

[therealglazou]

I would say "User agents commonly render unvisited hyperlinks" or "Visual user agents commonly display unvisited hyperlinks".

[ionas]

Now instead of messing with the spec and creating a dirty implementation that works in weird ways (like explained in dbarons blog) why could the standards body not simply remove/deprecate :visited or at least move it into a different security profile (such as trusted or local single page applications only).

Then to enable the use case (notify users of a link that has already been visited) browser vendors could implement a tool-tip that is rendered ontop of the canvas not reachable by getComputedStyle, PointerEvents or anything else as it is a OS-tooltip, which would somehow notify the user about the state of visit.

Now this is not the most convenient for users, but it is the most secure.

Another way around this would be security profiles where :visited would always work on urls of the same domain, maybe even across subdomains of the same domain, but never onto foreign domains.

Foreign domains could then implement something like access control origin where they declare which domains would show the users visiting history, users could then accept those shared history suggestions/lists.

While all these are just bits of ideas, I am pretty sure at some point in future dbarons implementation will bite back because some browser API want's to read what's the pure render color below the mouse pointer or whatnot. :visited security constraints will be forgotten quickly or ignored by "greater needs".

[tabatkins]

:visited already exists and is commonly used; dropping it entirely isn't realistic.

There've been discussions (off-list, I think) about figuring out ways to expose :visited styling "safely" in some contexts: allow it to work normally for same-origin links; or remember what links you actually clicked from a given domain, or visited from with a Referer header (and thus, what history could have been snooped on via JS or server code anyway), and let it work normally for those links, etc.

[dbaron]

Regarding "In full, the behavior is actually that the '':link'' pseudo-class applies to all links, ": I don't think "applies to all links" sounds right here. It might be better to restructure this somehow to say that ::link applies to all links that ::visited doesn't apply to.

(But it seems like this PR is pretty messed up... the "Files changed" view for the PR as a whole doesn't have the first commit in it... and trying to comment on the commit with the actual changes fails with "Start commit oid is not part of the pull request".)

[tabatkins]

Yeah, I have no clue what the PR is doing; I'll have to manually reconstruct it.

[dbaron]

(I was going to continue to review further... but I haven't actually read further yet.)

[astearns]

It looks like the substantive parts of this PR were merged in some other fashion