[PATCH] httprepo: use caps instead of between for compat check

[Augie Fackler]

On Mar 13, 2011, at 1:00 AM, Peter Arrenbrecht wrote:
> 
> On Sat, Mar 12, 2011 at 9:52 PM, Augie Fackler <durin42 at gmail.com> wrote:
>> On Mar 12, 2011, at 1:22 PM, Benoit Boissinot wrote:
>>> 
>>> On Sat, Mar 12, 2011 at 6:30 PM, Peter Arrenbrecht
>>> <peter.arrenbrecht at gmail.com> wrote:
>>>> # HG changeset patch
>>>> # User Peter Arrenbrecht <peter.arrenbrecht at gmail.com>
>>>> # Date 1299950942 -3600
>>>> httprepo: use caps instead of between for compat check
>>>> 
>>>> We usually query the capabilities at some point anyway so this way the first roundtrip
>>>> actually does useful work.
>>> 
>>> Queued, thanks!
>> 
>> What happens when the server responds with garbage that isn't a valid capabilities list? Do we abort with some silly unsupported capability? What would happen if we tried talking to a newer server and it required a capability we lack locally?
> 
> Maybe I'm missing something, but I don't think we would see anything
> different than today, since the capabilities are practically always
> requested at some point anyway. Even today, if the server responds
> with something silly, we would get a failure (because it would raise
> something other than RepoError).
> 
> The only problem I see would be if a static http server were to send
> garbage for the caps check, which it doesn't.
> -parren

Ah, I missed a key point: capabilities come back with a known content-type, which means it *is* a safe way to sniff for repo existence on modern servers.

I'm convinced it's safe after some cURL use, might send a doc patch to add some comments for future viewers if that seems desirable (it's nonobvious at first reading why it's acceptable to trust _fetchcaps() to raise RepoError IMO, but I'm also very tired at the moment...)

Augie