http transport should tell user if host is unknown rather than saying "not a branch"

[Michael Ellerman]

When pulling (and presumably doing other operations) over http, the non-existance of a host is reported as "not a branch", this can be rather confusing, eg:

concordia ~/test$ bzr pull http://bazzar-vcs.org/bzr/bzr.dev.as.knit
bzr: ERROR: Not a branch: http://bazzar-vcs.org/bzr/bzr.dev.as.knit/

The message sends me off to check that I got the path right, which I did, I then pull my hair out for five minutes before realising I've actually got the host name wrong!

[John A Meinel (jameinel)]

We now get ConnectionError, which is better, though the error is still slightly confusing.

Leaving this bug open, because I think we can do better, but decreasing its priority.

This is the current errors:
% bzr branch http://nohost/foo/bar
bzr: ERROR: Connection error: curl connection error (Success) on http://nohost/foo/bar/.bzr/branch-format
% bzr branch http+urllib://nohost/foo/bar
bzr: ERROR: Connection error: Error retrieving http+urllib://nohost/foo/bar/.bzr/branch-format: <urlopen error (-2, 'Name or service not known')> <urlopen error (-2, 'Name or service not known')>

[Vincent Ladeuil (vila)]

Will fix that ASAP.

[John A Meinel (jameinel)]

The exception is a little bit ugly right now, but it does indeed say "Couldn't resolve host".

Considering this fixed, though vila may have futher improvements to make.

[Vincent Ladeuil (vila)]

This is what bzr.urllib.keepalive gives:

bzr branch http+urllib://nohost/foo/bar
bzr: ERROR: Connection error: Couldn't resolve host 'nohost' (-2, 'Name or service not known')
vila:~/tmp :( $ bzr branch http+pycurl://nohost/foo/bar
bzr: ERROR: Connection error: curl connection error (Couldn't resolve host 'nohost')
on http://nohost/foo/bar/.bzr/branch-format
vila:~/tmp :( $

I don't see the point of repeating the url requested as it appears just one line above, but still mentionning the original error may help in obscure cases.

Is it satisfying enough ?

[John A Meinel (jameinel)]

I think it is good to repeat the url. For one thing it is an explicit path (.bzr/branch-format) in case that is important in debugging things.

Also, it help in case we ever had something like aliases, so people do:

bzr branch foo/bar

Which is secretly aliased to http://foox/bar

I'm happy with it.