Hi,
Does anyone have any ideas why these tests are failing (e.g. http://kegel.com/wine/patchwatcher/results/3644.log)? The patches in question don't touch that code at all.
Looking at http://test.winehq.org (e.g. http://test.winehq.org/data/07fafa26a318a3da3185fcac71bfa6d3fba82787/xp_fg-w...) there seems to be some variance on WinXP and 2003.
Does anyone know if this is context dependant (e.g. network timing out)?
Thanks, - Reece
On Tuesday 28 October 2008 09:10:33 Reece Dunn wrote:
Does anyone have any ideas why these tests are failing (e.g. http://kegel.com/wine/patchwatcher/results/3644.log)? The patches in question don't touch that code at all.
Looking at http://test.winehq.org (e.g. http://test.winehq.org/data/07fafa26a318a3da3185fcac71bfa6d3fba82787/xp_fg-w...) there seems to be some variance on WinXP and 2003.
Does anyone know if this is context dependant (e.g. network timing out)?
Windows caches connections up to a couple of seconds, that's why you don't see notifications for hostname resolution or the connection being made.
I guess we could increase sleep between two tests that query the same host once more, or just pick another host.
-Hans
On Tue, Oct 28, 2008 at 1:37 AM, Hans Leidekker hans@codeweavers.com wrote:
... I guess we could increase sleep between two tests that query the same host once more, or just pick another host.
Any test that uses the public internet is buggy and is bound to fail often enough to annoy people when run as part of an automated test system; that's http://bugs.winehq.org/show_bug.cgi?id=15716 I suspect that's what's happening here.
Since only automated test systems run these tests, we could have an optional shell script, run manually on systems that have to run the tests really reliably, that sets up a local server and adds entries to /etc/hosts. Or we could implement our own mock servers as part of the tests in question.
Hans, since it's your code that's failing here, can you think of anything we can do to make the test pass even if the public internet is going up and down like a bounce house? Skipping the test if the first use of the internet fails isn't good enough, since the internet might go up and down during the test several times. - Dan
On Tuesday 28 October 2008 14:02:15 Dan Kegel wrote:
I guess we could increase sleep between two tests that query the same host once more, or just pick another host.
Any test that uses the public internet is buggy and is bound to fail often enough to annoy people when run as part of an automated test system; that's http://bugs.winehq.org/show_bug.cgi?id=15716 I suspect that's what's happening here.
Funny thing is that as we speak codeweavers.com is being slashdotted and some of the wininet and winhttp test fail because of that ;)
But I don't think we want to give up on this type of test coverage though, precisely because it exercises failure paths better, in tests and core Wine code.
(This particular test failure is not one of those BTW, its a cache effect that you would also see when testing a local server.)
Since only automated test systems run these tests, we could have an optional shell script, run manually on systems that have to run the tests really reliably, that sets up a local server and adds entries to /etc/hosts.
I'm all for this option; it would give more control over the tests, and being able to inspect a server log file is invaluable too.
-Hans
On Tue, Oct 28, 2008 at 8:22 AM, Hans Leidekker hans@codeweavers.com wrote:
But I don't think we want to give up on this type of test coverage though, precisely because it exercises failure paths better, in tests and core Wine code.
I agree. We just need to move the servers onto localhost.
-
Dan Kegel -
Hans Leidekker -
Reece Dunn