Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!hplabs!hpfcso!kah
From: kah@hpfcso.HP.COM (Kathy Harris)
Newsgroups: comp.sys.hp
Subject: Re: HP Customer Support...
Message-ID: <7370051@hpfcso.HP.COM>
Date: 4 Dec 89 18:57:36 GMT
References: <203@limbo.Intuitive.Com>
Organization: Hewlett-Packard, Fort Collins, CO, USA
Lines: 33


>I can recall in particular a bug I submitted against the TCP/IP
>sockets implementation on HP-UX 5.2 (I think) on an early 300 series
>machine.  With the aid of a few other engineers we resolved that when
>opened using "fdopen" to have two streams, one for reading and one
>for writing, the file I/O stream buffer size and the TCP/IP window
>size conflicted so that I would get garbage if I were trying to 
>push lots of information through quickly.  We pinned that baby down
>to the size of the buffers, had a nice small sample program that
>demonstrated the problem, AND figured out a workaround (using the
>setvbuf() command, if I recall correctly).  I can well recall that
>it was almost two months before I heard back from the Fort Collins
>Networking group, with the response that "it didn't fail on our
>machine, sorry."   That was it.  

Well, I'm sorry Dave but your info or memory about bugs not getting
resolved is just not correct, at least in this case.  This one was
handed to the person in charge of stdio.  It was resolved with the
following info:

	Stdio does not support two open streams connected to the
	same underlying file descriptor.  The correct way to
	set up code like this is to dup(2) the file descriptor,
	and then do an 'fdopen' on each descriptor. 
	
The networking group made this change and the code then worked fine.

Maybe you didn't submit the bug report, but instead someone else did,
so the resolution info went back to them instead of you.

Kathy Harris
Colorado Language Labs
Ft. Collins, Co.