On Thu, 14 Sep 2006, Michael Wallner wrote:

> I can confirm that the test program works fine now.
> Unfortunately my test case is still broken, not as bad as before, though.

So what's the difference between the test program and your test case? Can we
modify the test program to show the same problems?

> I wonder if it might be somehow related to "Pipe broke" debug messages? Note
> that this test is still run without pipelining enabled.

Yes, I think those outputs are a sympthom of badness.

> BTW, is a connection id of "-1" usual?

Yes, when doing many transfers. The "connection id" in that sense is simply an
index in to the connect cache/array and it currently only fits 10 connections.

An interesting test would be if you modified lib/url.c:Curl_mk_connc() to make
a connection cache with more entries than you do transfers, to see if that is
the culprit of these problems (which I suspect).

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html