Hi,
we dig into this subject further and dicovered that the problem is only in case of sending http header: Expect: 100-continue

Thats because the server will send reply: HTTP/1.1 100 Continue before the communication ends, and curl will assume that the server started sending data, which is false.

When HTTP/1.1 100 Continue is sent by server, curl will continue to sending real POST data. The counter should start when curl finish sending data, but it has already been set while 100 Continue was recived.

This is why the counter STARTTRANSFER is bugged - in this particular situation, when curl uses Expect: 100-continue

We will try to handle his and send the patch.
Best regards,
Mezon

---
** [bugs:#1213] Wrong STARTTRANSFER timer accounting for POST requests**
**Status:** closed-later
**Labels:** POST CURLINFO_STARTTRANSFER_TIME 
**Created:** Sat Apr 06, 2013 07:17 PM UTC by Mezon
**Last Updated:** Sat Mar 22, 2014 09:01 PM UTC
**Owner:** Daniel Stenberg
curl 7.28.1 (i686-pc-linux-gnu) libcurl/7.28.1 OpenSSL/0.9.8o zlib/1.2.3.4
Protocols: dict file ftp ftps gopher http https imap imaps pop3 pop3s rtsp smtp smtps telnet tftp 
Features: IPv6 Largefile NTLM NTLM_WB SSL libz 
There is an error in timing accounting at least for STARTTRANSFER timer during POST requests.
Timer works fine with GET requests, but while using POST the time for CURLINFO_STARTTRANSFER_TIME is wrong. While using POST CURLINFO_STARTTRANSFER_TIME minus CURLINFO_PRETRANSFER_TIME is near to zero every time.
Proof of concept:
I've prepared CGI script that sleeps 3 seconds on start, before it returns any data.
While making curl GET request to the script, (CURLINFO_STARTTRANSFER_TIME - CURLINFO_PRETRANSFER_TIME) is ok [~ 3 seconds]. Trying the same thing with POST request, cause the 3 seconds appears in CURLINFO_TOTAL_TIME instead of CURLINFO_STARTTRANSFER_TIME. So with POST it looks like the transfer took 3 seconds instead of waiting for server reply.
IMHO the problem is in ./lib/transfer.c in readwrite_upload() function, where Curl_pgrsTime(data, TIMER_STARTTRANSFER) is called in wrong place. Unfortunately I'm not sure where to replace it to fix the bug. I think it should go somewhere in Curl_readwrite() function, for example here:
~~~~~~
/* We go ahead and do a read if we have a readable socket or if
     the stream was rewound (in which case we have data in a
     buffer) */
if((k->keepon & KEEP_RECV) &&
     ((select_res & CURL_CSELECT_IN) || conn->bits.stream_was_rewound)) {
    result = readwrite_data(data, conn, k, &didwhat, done);
    if(result || *done)
      return result;
  }
  else {
      Curl_pgrsTime(data, TIMER_STARTTRANSFER); // <-- here
  }
~~~~~~
I'm not sure if any other timers are affected.
Best regards,
Mezon
---
Sent from sourceforge.net because curl-tracker@cool.haxx.se is subscribed to https://sourceforge.net/p/curl/bugs/
To unsubscribe from further messages, a project admin can change settings at https://sourceforge.net/p/curl/admin/bugs/options.  Or, if this is a mailing list, you can unsubscribe from the mailing list.