On Mon, 18 Mar 2002 John.Clayton_at_barclayscapital.com wrote:

> Sometimes, our apps are unable to complete their curl library socket()
> calls. The error occurs within connect.c of the curl library, on line 594
> (version 7.9.4). The error we're getting is WSAEPROTONOSUPPORT, running on
> Windows NT SP3.

I would feel safer if you took the step up to 7.9.5 first. Many bugs were
ironed out in that release.

What bothers me about this problem is your first word here: "Sometimes". If
this was a problem with the input parameters to the socket() call, it should
fail every time not somtimes...

> We form up HTTP requests, and always spit XML out from our servers - which
> is received and processed by Curl and then fed into James Clarke's expat
> library (anyone know interop. issues with these two libs?).

I can't think of any reason why they shouldn't play nice with each other. Not
that I have any in-depth knowledge of expat.

You say "sometimes". How often is that? When the problem occurs, do the
subsequent calls usually also fail? Hoe many calls normally succeed before it
fails?

> - We cut/paste of our production code into an existing test harness which
> loops across 2000 times with one of our more common queries - no problems
> found.

So what differs in the test environment compared to where you see this
problem?

> - Changed the Curl code to check the result of the create socket() call -
> here's what our connect.c @ line 459, using curl 7.9.4 - this is the
> location of the failure - but changing to PF_INET didn't help

Hm, what about a lame suggestion. What if you would sleep of a little while a
retry the exact same socket() call a number of times to see if waiting a
little might solve it?

My reasoning here is quite frankly that something is rotten in the socket()
implementation...

I'm sorry I can't be more specific. This sounds like a tough one.

-- 
    Daniel Stenberg -- curl groks URLs -- http://curl.haxx.se/