The Jelmer "fork" of bzr.dev

[Jelmer Vernooij]

Forest Bond wrote:
> Hi,
>
> On Sun, Apr 12, 2009 at 02:14:20AM +0800, Eugene Wee wrote:
>   
>> On Sun, Apr 12, 2009 at 2:07 AM, Russel Winder
>> <russel.winder at concertant.com> wrote:
>>     
>>> More seriously, I think the behaviour really is counter-intuitive for
>>> the user and so I think the problem does deserve some workaround.  I
>>> agree the whole idea of having vendor/website specific exception is
>>> anathema, but if that makes the user experience better then perhaps it
>>> is worth considering.
>>>       
>> Would it be feasible to call it a bug with Google code hosting and
>> request that this be fixed on Google's side?
>>     
>
> I don't think it's a bug to return 401 for a request to a service that requires
> authentication.  It's unfortunate that you have to authenticate only to find out
> that you are using the wrong protocol, but this is a consequence of using the
> same URI schema for multiple protocols.  So it goes.
>   
I'm not sure if POST requests actually require authentication or whether
they are simply not supported (in which case 405 seems a more
appropriate response).
> If bzr could only ask the user for authentication info once and then use it for
> both the original bzr-native request and then retain it for use with the bzr-svn
> request, that would be ideal.
Bazaar will already ask for the credentials only once if credentials are
asked. The problem in this particular situation is that GET and OPTIONS
requests (used by bzr-svn) don't require authentication while POST
requests (tried by bzr's smart protocol client) do. We don't want Bazaar
to ask for credentials here at all, but as the smart client gets to
probe the URL before bzr-svn does the user will see a prompt.

Perhaps Bazaar should try all formats without using authentication first
and if there's no format that succeeded prompt the user and try the
formats that required authentication. That's quite a lot of
restructuring to implement properly though :-/

Cheers,

Jelmer