cURL / Mailing Lists / curl-library / Single Mail

curl-library

Re: Serious sporadic bug in libcurl 7.10

From: Gary Lawrence Murphy <garym_at_teledyn.com>
Date: Fri, 10 Jun 2005 12:02:56 -0400

>>>>> "R" == Ralph Mitchell <ralphmitchell_at_gmail.com> writes:

    R> Have you tried dropping a "sleep (5);" in at the end of the
    R> main loop?

Here's were I betray my C getting way too rusty from mis-use: Because
of the SIGALRM handler, I had dropped sleep() calls from the code --
the sleep(3) man page advises

    sleep() may be implemented using SIGALRM; mixing calls to alarm()
    and sleep() is a bad idea.

what it doesn't say is what I could use instead. But I love your
theory:

    R> I seem to remember reading somewhere that some kinds of network
    R> packets can be "expedited", thereby passing others destined for
    R> the same host. I think it was mostly a problem where the
    R> sender could tell the receiver "all done, drop the connection",
    R> and that message would overtake the last few data packets. The
    R> receiver would then miss those packets.

    R> Obviously that's not the problem here, but maybe the startup
    R> (SYN) packet is expedited the same way, so the PUT headers
    R> overtake the last data?? That's a wild, completely unsupported
    R> guess... :)

Makes a lot more sense than my time-travel software tachyonics theory :)

    R> Also, try a packet sniffer like ethereal.

Also very excellent advice; this would provide a far better smoking
gun than the socket-level monitoring. A thousand thanks for this.

-- 
Gary Lawrence Murphy <garym_at_teledyn.com> ==============================
www.teledyn.com - blog.teledyn.com - irish.teledyn.com - sbp.teledyn.com
====================== The present moment is a powerful goddess (Goethe)
Received on 2005-06-10