On Tue, Sep 04, 2012 at 11:41:50AM +0100, Frank Shearar wrote:
> Hi,
> 
> I meant to nag a lot more about the failing tests, and I haven't.
> Sorry for not being sufficiently annoying!
> 
> We still have a bunch of failing tests -
> 
> KernelTests.Chronology.TimeStampTest.testFromSeconds
> CollectionsTests.Streams.RWBinaryOrTextStreamTest.testExisiting
> KernelTests.Chronology.TimeStampTest.testReadFromA1
> NetworkTests.Kernel.SocketTest.testSendTimeout
> Tests.Localization.LocaleTest.testLocaleChanged
> Tests.Monticello.MCFileInTest.testStWriter
> Tests.Monticello.MCMczInstallerTest.testInstallFromFile
> Tests.Monticello.MCMczInstallerTest.testInstallFromStream
> Tests.Release.ReleaseTest.testNoObsoleteClasses
> Tests.Release.ReleaseTest.testUndeclared
> 
> - and I'd really like them to be green for a release. The Monticello
> tests cause noise, and so the ReleaseTests should pass if we don't run
> the MC tests. The SocketTest is, as I understand it, an issue specific
> to Linux, and doesn't occur on Window.

SocketTest>>testSendTimeout fails on my Linux system, although I cannot
say if it is a real failure or if it is due to a difference in the
way networking functions on unix versus windows. In any case, if it is
a real problem, it is certainly not a *new* problem, and it should not
stand in the way of an image release. But labeling as an expected
failure would not be right, because the test presumably will succeed
on many platforms.

On my Linux system, I also get a failure on SocketTest.testSocketReuse
because it tests for a feature that does not exist on this platform
(setting the socket option option 'SO_REUSEPORT'). Apparently this
option must be supported on our test platform, but FYI it is likely
that other users will see this "failure" when they run the tests.

> 
> If I _have_ to, I'll move the tests into "expected failures", and
> re-enable them when we start work on 4.5.
> 
> One big gap in our testing infrastructure is testing on non-Linux
> platforms. I would like a kind soul to take up CI-like duties and
> commit to run the entire suite on OS X and Windows reasonably often.
> Any takers? (Ideally we'd have build slaves for this, but...)
> 
> frank