Re: Address bar searches (was: epiphany-list gnome org)

["Joseph S. Huang" <josephshuang gmail com>]

I agree with you. That mozilla bug is fixed, so I don't know what the
person who filed Epiphany bug is talking about. It is quite vague.

If I enter a non-resolving domain with a dot, Epiphany gives an error
message. The current behavior seems to be to treat it as a URL if it
contains a dot. If it contains a colon and not a dot, then if the DNS
doesn't resolve, it strips off the colon, and treats it as a keyword.
The only problem I see is that if someone mispells an entry in their
hosts file, then it will be directed to google, with the port number
chopped off, if it included one. URLs that start with http:// seem to
always be treated correctly.

Also, how about an error page for DNS names that fail to resolve,
perhaps which could contain suggestions (closely matching previously
visited domains?) and let the user search for the term(s) entered. 

> A problem with that is that many entered strings that are meant to be
> URLs don't start with a protocol. You enter 'gnome.org' or
> 'localhost', not bothering with the protocol. The current workaround
> for several browsers seems to be to identify everything with a dot
> present as an URL, even when they are clearly not, as in normal
> sentences with trailing space after the dot, or a string ending in a
> dot. Even taking that into account, it means that me searching for
> 'document.pdf' fails because it look like an URL - and on some
> internal companies LAN or in the future it might be a valid URL,
> further complicating matters.
> 
> That's why I think it's intiutive to try DNS for everything that might
> be valid and if there's no hit, try a search instead. If the site is
> down, but exists in DNS, there is no search - the browser has already
> moved on to trying a fetch. If that fails, error messages are
> appropriate. If it doesn't exist in DNS, chances are that you meant a
> search (or that you are mistaken about the URL, in which case a search
> also might help). This goes for strings that *do* start with http://
> as well.
> 
> If you do surf sites that, if misspelled or removed from DNS, are so
> sensitive that it would be a real concern to have them forwarded to a
> search engine, then by all means it should be possible to disable. I
> don't want to deny anyone that right, I just don't see how that is
> even remotely the common case.