cURL / Mailing Lists / curl-library / Single Mail


test suite on windows (was Re: HTTPS via schannel not working with multi-interface)

From: Daniel Stenberg <>
Date: Mon, 9 Jul 2012 16:01:27 +0200 (CEST)

On Mon, 9 Jul 2012, Guenter wrote:

(I changed the subject to fit the topic better.)

> I recently tried to run the testsuite native on Windows through MSYS
> .........
> it was more than a pain: it lasted more than 2.5 hours, and tons of
> failures,

I think it is great that you actually reach that far! It means there's hope
and that we should work on getting the remaining problems fixed.

I won't rule out that the test failure are indeed legitimate failures. I took
a look at the first one in your output, test 46, and that looks like a mistake
in the test as it relies on a sort order from qsort() being the same when the
sort key is identical which I believe is wrong. I posted a change just now
that I hope will make the test more reliable. Possibly more of the test
failures are in this style. 62 looks like it is the same cause.

Test 1105 seems to be a case of CRLF line endings vs LF line endings so the
test might simply need a minor edit to go through.

Test 162 fails because your curl build sends different NTLM data than the test
expects, but since you build with SSPI it isn't that strange really. We need
to either extend the test to either:

  1 - cut off the string at some point before doing the comparison, which is
      something we do for some tests already but takes away some level of
      verification that is nice to keep...

  2 - introduce alternatives in the format so that it could check for A or B to
      match somehow

  3 - make the test depend on features no otherwise present. Like this test
      could be made to rely on "!SSPI" perhaps so that it would only run if not
      SSPI is a feature, and then we could make a clone of test 162 that would
      rely on SSPI being _present_

  4 - something else

I think I personally favour alternative 3. "Unfortunately" we do a whole bunch
of NTLM tests which would mean a whole bunch of cloned and modified test cases
to edit.

> mostly path-related with the test results; and finally it got hung at one
> point ...; also now unfortunately the ftp test server does no longer work on
> Windows at all ...

I don't know of any particularly good reason why it isn't! Do you?

List admin:
Received on 2012-07-09