On Mon, 2 Mar 2009, Brian J. Murrell wrote:

>> Right. I think the logic that tries both versions should rather be moved
>> into the Curl_input_negotiate() function and that could then avoid doing
>> that fail output while "probing".
>
> Yeah, I wondered that too, but it then becomes unclear to me whether this
> continued distinction between Negotiate and GSS-Negotiate is needed outside
> of Curl_input_negotiate() or whether we can isolate the need to distinguish
> between them to Curl_input_negotiate() and collapse all uses of
> "GSS-Negotiate" into "Negotiate" (i.e. string replace) outside of
> Curl_input_negotiate().

I checked just now and I can't spot any particular distinction at all made
outside of http_negotiate.c in the current code. I think we should strive
towards keeping this tradition and make the code outside of that file treat
Negotiate and GSS-Negotiate as if they truly are the same thing.

> Is there really any need to pass Curl_input_negotiate() anything other than
> a single "Negotiate" value? In fact is there any need to pass it any value
> anymore given that we are always going to try both to see which one is in
> use?

It still needs to pass on the connect data struct for the connection, and tell
the function if it wants server or proxy auth. The 'header' parameter could be
used to not "probe" for the service in case the exact method is known. Keeping
the exact same prototype of course also keeps the amount of changes smaller.

-- 
  / daniel.haxx.se