Hi dan,

   Sorry to puzzle you with unreadable response. But I formatted it well when I sent it.

   Also, since we have discussed it here, is it necessary to change to curl-library mailing list?

> So "neg_ctx->output_token.value" was NULL at that time?
> 
> If so, we should add a check before the Curl_base64_encode() is called there 
> and bail out if it is. Right?

    Well, both neg_ctx->output_token.value and neg_ctx->output_token.length are zero at that time.

    Maybe we should add a check before the Curl_base64_encode(), but I think even Curl_output_negotiate needn't be called if neg_ctx is filled by zero.


> > It's true that NTLM authenticates the connection, but it depends on the 
> > challenge string sent from server.
> 
> Depends how? If I understand you correctly you say that NTLM can still be 
> negotated for you even if the server drops the connection during the 
> handshake. I've personally never seen that happen and I didn't know it was 
> even possible - for the reason already stated that NTLM authenticates the 
> connection and thus disconnecting makes it strange since then it isn't longer 
> the same connection for which the autentication started for.
    Yes, the handshake is done by http protocol, and http protocol was designed as a nonpersistent protocol, so I think server is entitled to close the connection at each response.

    I'm just a beginner of NTLM, so I don't know if there's some special rules for http connection when using NTLM. But I think NTLM can still be negotiated even if the server drops the connection after response. 

> Can you show us your work-around for this problem as a diff -u output? Or even 
> better, have you thought about or produced a proper fix for it? We have made 
> some minor NTLM fixes in the current code, so you may want to base any 
> efforts, tests or further patches on the CVS version and not 7.18.2...
    For my current work, I'v changed a http_ntlm.c to fix this, but since I haven't read most src of Curl, my fix may be incondite. My work-around is fedora8+websphere6.1+curl 7.15.3, also I'd tested on 7.18.2 and get the same result. Websphere6.1 will close connection each time ,may be you can try to download a trial version and test it.

     I should say sorry that I can't understand the meaning of "show you my work-around for this problem as a diff -u output", but I'll send you a diff file between my fix and http_ntlm.c of 7.15.3. I'll also send you the outputs of curl -v. 

     All this will be appended as attechment in next mail ,I'll send it after about 9 hours.