[csw-maintainers] apr-1.3.9 - test failure ?

[Sebastian Kayser]

rupert THURNER wrote on 22.12.2009 01:27:
> On Tue, Dec 22, 2009 at 00:16, Sebastian Kayser <skayser at opencsw.org> wrote:
>> rupert THURNER wrote on 22.12.2009 00:04:
>>> i commented the tests for the moment, the test code does ipv6.
>>
>> glad it helped! There was a similar problem with mbuffer, which also
>> does IPv6 tests. Instead of completely disabling the test suite, it was
>> possible to only skip the IPv6 test. This way one is still alerted (and
>> I already was) if anything else goes wrong. Might be wise to take the
>> same route for apr for future builds. Do any other tests fail right now?
>>
> no. i did not find any log file but looked into the code. i do not
> understand what they really try to do ... but i'd saythe test suite
> should not test things which are known not to work.

Regarding the test suite: There is a bug report about the IPv6 testing
problem [1] where the bug reporter would also like to see the IPv6 tests
being skipped rather than failed for boxes without IPv6 interfaces. The
apr guys take a different stance.

Luckily, a configured IPv6 loopback interface is sufficient. @Buildfarm
admins: would it be possible to configure IPv6 loopback interfaces for
the buildfarm boxes?

# ifconfig lo0 inet6 plumb
# ifconfig lo0 inet6 ::1 up
# touch /etc/hostname6.lo0

> the code contains:
> #if APR_HAVE_IPV6
> static void tcp6_socket(abts_case *tc, void *data)
> 
> and the config log says:
> configure:34230: checking if APR supports IPv6
> configure:34232: result: yes

That's something different. ./configure detects general IPv6 support in
the OS and it's libraries and thus build apr with IPv6 support. It is
perfectly fine to build apr with IPv6 support - even if the particular
build box doesn't have a configured IPv6 interface. This way we end up
with an apr package that our users can use, even when they are running
it in a configured IPv6 environment.

Sebastian

[1] https://issues.apache.org/bugzilla/show_bug.cgi?id=45494