Bugzilla

> (1) Does the current patch upgrade all failed connections if [mAllowKeywordFixup is true](https://hg.mozilla.org/try/file/b2f63690cb3e77378491b64d54a03cebd88934c4/docshell/base/nsDocShell.cpp#l6039)? 

Yes, I was attempting to upgrade as much as possible when port 80 connections are rejected.  Maybe that's a mistake.  I'm still confused, however, about why `localhost:80` doesn't serve data for the mochitests.

> (2) If above does not apply and these tests just want their error page, they should either
> a) disable a `browser.fixup.fallback-to-https` pref before running or
> b) be changed to a hard-coded domain that won't be upgraded, e.g. http://offline.invalid/.
> 
> A dedicated pref might anyway be useful for debugging.

This seems reasonable, but I'd hate to add more complexity to the browser if we don't need to.  Maybe I'll make another patch for the pref just in case.  Thanks for the idea!

> (1) Does the current patch upgrade all failed connections if [mAllowKeywordFixup is true](https://hg.mozilla.org/try/file/b2f63690cb3e77378491b64d54a03cebd88934c4/docshell/base/nsDocShell.cpp#l6039)? 

Yes, I was attempting to upgrade as much as possible when port 80 connections are rejected.  Maybe that's a mistake, but the test failures are all top-level URLs (you can see them in the url bar as the test runs).  I'm still confused, however, about why `localhost:80` doesn't serve data for the mochitests.

> (2) If above does not apply and these tests just want their error page, they should either
> a) disable a `browser.fixup.fallback-to-https` pref before running or
> b) be changed to a hard-coded domain that won't be upgraded, e.g. http://offline.invalid/.
> 
> A dedicated pref might anyway be useful for debugging.

This seems reasonable, but I'd hate to add more complexity to the browser if we don't need to.  Maybe I'll make another patch for the pref just in case.  Thanks for the idea!