It's not about history, it's about TYPED history. if you type even just once the https version, it should go to the https version regardless of http being in history before or after it. 42 or 43 shouldn't make a difference. Doesn't this work?

Ok, I was wrong in comment 3 (obsoleting it). In bug 769348 we changed the algorithm to figure out the best schema for an host, we decided to go with "use a prefix if all typed pages for that host use that prefix". That was after a long discussion in bug 769994 where users needed a way to override our too aggressive algorithm that was forcing a prefix even if just one entry had that prefix. Let's make an example, supposing the host "mozilla.org". Before bug 769348 it was enough that just one typed page with host "mozilla.org" had "https" schema, for us to suggest an https autofill entry. That was nice from a security point of view, but opened a can of worms (bug 769994), mostly due to https pages returning errors or broken certificates and other situations we cannot detect in Places. So it was decided in bug 769348 to suggest a given prefix for "mozilla.org" only if all typed pages have that prefix. The second solution offers a way out to users that couldn't always use https on an host. On the other side makes extremely easy to lost secure prefix just by typing once the non secure host. Typing in reality includes various user actions: - Opening a tab from "another device" (Sync) - middle-mouse paste - typing an url - loading one (or multiple) pages from any history view On a second thought, I think both decisions were pretty extreme, what matters is that we should somehow be able to adapt to the user/page changes over time. I think we have 2 problems here. The first problem is that we are abusing the typed flag and that pollutes data. It's not even anymore a real typed flag, since clicking on any history view sets it, plus we don't set it for loads from the bookmarks views that are likely even more important? Due to this polluting, since autofill is based on typed being set, we autofill a page loaded once from history, but not a bookmarked page :( This is sub-optimal. We could try to restore the typed flag to its original intent, remove browser.urlbar.autoFill.typed (it's the default) and change the algorithm to decide what to autofill. We decided to autofill only typed domains in bug 720258, since otherwise we were autofilling pages that the user visited just once... Using typed looked like a simple solution, but it's again sub-optimal. A better decision may be based on frecency of the host, we could autofill only if it's over a given threshold (TBD). This would likely also bring some perf wins, since keeping typed in sync has a cost. The second problem is that by stating "use a schema only if all typed pages have it" does not adapt with time. it's enough to have one single wrong visit to break the profile forever. Again, I think a better solution would be to base the decision on frecency, like we could say "use a schema only if all of the 3 most frecent pages have it". Both algorithms can be tweaked by changing the frecency threshold and number of frecent pages. The work to implement these can be splitted into 3 to 4 bugs, it's not extremely complex but involves changes to the awesomebar behavior, imo they will end up being improvements, since we will be able to better adapt to each single user, but it's still touching the most common point of interaction and autofill behavior. I'd like to get some thoughts on these suggested changes. Plus, I'd like to know from managers whether it's worth I spend some time on trying to implement them, these affect awesomebar quality.