This, and some other recent CI failures, were due to initialization
ordering. Works for me locally, did not work on Jenkins. Order matters.
Assembler is an interface. It also has calculated constants. But these
constants are not assigned inside class initialization sequences because
class/interface initialization does not happen when classes are loaded
due to other initializations.
Being an interface, it's hard to hook in and get the setting of
Assembler.general to be done before ARQ or TDB get touched.
Some tests
The good news is that the random test order jena-permissions flushed
this out. It is nothing to do with jena-permissions code, only the
random order that its test run.
ARQ in maven runs random order as well - but didn't stumble on the
problem. It takes direct and early mention of Assembler to cause the
problems.
Windows test CI failures are different - fails in timeout tests (exactly
which moves around). Don't know what to do about these - just longer
timeouts hasn't worked. Really long timeouts will delay regular
testing. It seems to be when the CI slave is running another job in
parallel so some weird time slicing issue maybe?
Andy
-------- Forwarded Message --------
Subject: Jenkins build became unstable: Jena_Development_Test #2089
Date: Thu, 24 Sep 2015 13:50:48 +0000 (UTC)
From: Apache Jenkins Server <[email protected]>
Reply-To: [email protected]
To: [email protected]
See <https://builds.apache.org/job/Jena_Development_Test/2089/changes>
