[wtp-releng] Results of investigating massive JUnit failures

[David M Williams <david_williams@xxxxxxxxxx>]

For the curious, by "trial and error"
I narrowed down the cause of the massive JUnit failures removing an old
version (1.7.0) of org.apache.ant from the releng test feature (while trying
to cleanup stuff we no longer used). 

I've added it back for now, and the more
recent prereqs, such as for 0414 platform. 

I've opened bug 309621 to track the issues
until we can remove this old bundle. 
https://bugs.eclipse.org/bugs/show_bug.cgi?id=309621
I don't understand it all, exactly, but
think its presence (or absence) changes the exact order and "wiring"

between various classloaders. 


I do think this is complicated by us having
"mixed" constraints on JUnit 4, so I cleaned up some of that,
as much as I could. 
This means some junits might fail, and
need "real" attention, instead of me just changing the manifest.mf
files. 


... we still have some bundles that require a
narrow range of JUnit, e.g. 
Require-Bundle: ...
 org.junit;bundle-version="[3.8.0,4.0.0)"

Specifically, 
org.eclipse.jst.server.generic.tests
org.eclipse.wst.internet.cache.tests
org.eclipse.wst.xml.validation.tests [this one was fixed in head, but not
released]
org.eclipse.wst.xsl.exslt.ui.tests
org.eclipse.wst.server.http.core.tests

The first 4 I could fix in manifest.mf and there were no compile errors,
but of course I don't
know for sure they will still run ok. 

the http.core.tests one needed some "quick fixes" to fix compile
errors ... so
even a greater chance that it won't run (test) correctly. 

But, lastly, I discovered that 
org.apache.commons.discovery
from Orbit still had a narrow range for (optional) JUnit bundle. 
See bug
309619.
I've fixed in Orbit, but will take a few days to trickle through to where
we
can include it. 

At that time, it'd be worth trying again to remove our org.apache.ant version
1.7.0
again.

I hope you all had a fun weekend too. :)