cURL / Mailing Lists / curl-library / Single Mail


Re: test513 failure on s390

From: Kamil Dudka <>
Date: Thu, 3 Dec 2009 23:49:00 +0100

On Thursday 03 of December 2009 23:04:19 Daniel Stenberg wrote:
> That's certainly curious. And isn't server.input looking pretty much like
> that POST request it wanted?

Yes, I can only see one more newline at end of the expected input in contrast
to the dump of server.input.

> All file checks that does between what a test should get and
> what it gets is run through a function and that function generates
> log/check-expected to look like what to expect and log/check-generated to
> contain what was just gotten. They do variable expansion and so on. If the
> two files are not equal, something is wrong.

Looking at the diff output it looks like log/check-generated is completely

> The same kind of check is done for several different things, so the
> information showed just before on the test output provides the key for what
> was actually tested when it failed to match.

Can I somehow increase verbosity of the testing framework itself?

> No, it's taken from server.input which is the raw data sent the test HTTP
> server writes down to show the full request. But that's just equally
> strange as from what I understand on your description the server.input
> files looked the same in both the working and the non-working case.

I've just modified libtest/lib513.c to return wrong error code. This forces
the test-suite to write out all logs ... and yes, the content of server.input
is equal in both cases. The output of server.response as well, not counting
the PID of server.

> No, I too have a hard time to see how this happens... These cases tend to
> be very hard without actually being able to login and run and debug these
> things hands-on. They can in fact still be tricky even then!

Well, there is a chance I get a SSH access to a s390 machine. Do have any
suggestion what to try in that case?

Thanks in advance!

List admin:
Received on 2009-12-03