Hi Daniel,

On Fri July 30 2010 01:12:33 Daniel Stenberg wrote:
> I've pushed my edits of Kamil's original work, to let everyone able to test
> this and to work out the quirks more easily.
>
> We now test NTLM much better by faking a fixed host name when the test
> suite is running. This way we can be much safer that the different crypto
> backend glue codes are doing the exact same operations and thus work
> identically.
>
> The system is meant to make the NTLM tests run fine on systems that can do
> LD_PRELOAD _or_ build a debug-version of libcurl. All other systems will
> not be able to run the NTLM tests.

sorry for the delay, now I finally got to test it. I think that something
went wrong in handling of environment variables along with the push of
b5c3fed.

After each test-case the variables set in <setenv> should be thrown away, even
if the test is skipped. However there seems to be an interference on
$http_proxy among test63 and test67. If test63 is skipped, the http_proxy
variable from there is taken by test67 and causes it to fail unintentionally.
The problem can be repeated by:

./runtests.pl 'HTTP NTLM auth'

The other problem could be, that the NTLM tests override any LD_PRELOAD
variable given by user. The best would be to do something like:

LD_PRELOAD="./libtest/.libs/libhostname.so:$LD_PRELOAD"

... but the perl script doesn't take it. Any idea how to deal with it?
Unfortunately my perl skills are not good enough to help here :-/

Kamil
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2010-08-11