On Mon, Sep 7, 2015 at 12:33 PM, Bert Huijben <b...@qqmail.nl> wrote: > > >> -----Original Message----- >> From: Greg Stein [mailto:gst...@gmail.com] >> Sent: maandag 7 september 2015 11:29 >> To: Ivan Zhakov <i...@visualsvn.com> >> Cc: dev@serf.apache.org >> Subject: Re: [serf-dev] [serf] r2489 committed - In preparation of serf >> 1.4.0, >> remove the get_remaining function from t... >> >> How about: stop the debate? Remove it for 1.4.x (done), and bring it back >> for 1.5.x? ... The stuff I was doing on trunk for 2.x, I will move to a >> branch. Trunk will be 1.x. >> >> There is no reason to keep get_remaining in 1.4. It sounds like we have >> further work to keep compatibility regardless. Some Windows linkage thing, >> that I don't entirely understand. >> >> But let's move forward, rather than complain. There is a 1.4.x release to >> get out. > > There are still two test errors on at least two major platforms: > Windows reports > 1) test_ssl_missing_client_certificate: > /srv/subversion/buildbot/slave/serf-x64-macosx/build/test/test_ssl.c:1815: > expected <120172> but was <120199>
This test works most of the time, but will fail on slower systems, or with logging enabled. I think it's caused by an issue in the test framework but I doubt this test is really useful. What it tests, is that serf handles an SSL Alert with the correct error code. The SSL Alert is normally sent by the test server when it expects a client certificate, but the test explicitly doesn't send one. The test fails when the server didn't have the chance to send the alert bytes, but for that it gets only one chance, as the server loop isn't repeated as soon as the test framework reports an error. We can disable this test for now. > 2) test_ssl_server_cert_with_san_and_empty_cb: > /srv/subversion/buildbot/slave/serf-x64-macosx/build/test/test_util.c:541: > expected <0> but was <120199> > (See > https://ci.apache.org/builders/serf-x64-macosx/builds/6/steps/Test/logs/stdio) > > And Mac reports with openssl 0.9.8 > 1) test_ssl_server_cert_with_san_and_empty_cb: > /srv/subversion/buildbot/slave/serf-x64-macosx-default-openssl/build/test/test_util.c:541: > expected <0> but was <120199> > (See > https://ci.apache.org/builders/serf-x64-macosx-default-openssl/builds/1/steps/Test/logs/stdio) I managed to fix this test by recreating all certificates using the create_certs.py script. Not sure what changes though, the certificate used by this test is very simple (just like the others). Worth looking into a bit deeper. > > And with a more recent openssl: > 1) test_ssl_server_cert_with_san_and_empty_cb: > /srv/subversion/buildbot/slave/serf-x64-macosx-apr1_5/build/test/test_util.c:541: > expected <0> but was <120199> > 2) test_ssl_renegotiate: > /srv/subversion/buildbot/slave/serf-x64-macosx-apr1_5/build/test/test_ssl.c:1771: > expected <0> but was <120199> > (See > https://ci.apache.org/builders/serf-x64-macosx-apr1.5/builds/5/steps/Test/logs/stdio) > > I think I've had these errors on Windows for at least 3 months, probably > longer. > (120199 is a generic serf test failure. The cases I looked at return this > from the test http server) > > > I have a hard time understanding these tests and I don't know what it would > take to resolve these. > Enabling logging in serf and the test framework would be a good start I guess. > > Bert > Lieven
