Dain Sundstrom wrote:
There is a lot of text here and I'm a bit confused. FWIU, to use the
trick we used in bootstrap you need the following:
1) add your endorsed jars to the manifest class path
2) no code accessed during the boot process can reference an endorsed
package
3) modify the endorsed system property
4) use reflection to load a class that accesses the endorsed class
(this may have to happen in a child class loader)
That's not the behavior I'm seeing in any of the experiments I've seen.
The directories searched for endorsed appear to be fixed at JVM boot
time, so Geronimo's additions to java.endorsed.dirs in Daemon were
having no effect. I tested this with both the yoko-spec-corba jar and
the Xerces XML specs. In both cases, I was getting the JVM native
classes. In order to get this working correctly, I set GERONIMO_OPTS to
-Djava.endorsed.dirs=C:/..../lib/endorsed. Once that was set, the
correct classes were getting returned. I saw the same behavior from
some simple tests I coded up.
Rick
I have high confidence in this process due to the tck passing.
Additionally, I received a customer complaint that we were overriding
the xml processor on Java5 with the older spec version included with
Geronimo. Now it may turn out that this magic only works with the xml
specs, but I'd be very surprised.
-dain
On Sep 27, 2006, at 8:11 AM, Rick McGuire wrote:
I've been wrestling with a Yoko ORB problem when running under Java
5. When running on 1.4.2, the Yoko ORB works fine because the native
JVM doesn't include an implementation of
org.omg.PortableInterceptor.IORInterceptor_3_0. Because there's no
conflict, the yoko version is getting loaded.
With Sun's Java 5 impl, there is a version of
org.omg.PortableInterceptor.IORInterceptor_3_0 that's incompatible
with the CORBA standard (and also the Yoko implementation). The Sun
version is getting picked up, cause ORB initialization failures.
This is occurring even though the yoko-spec-corba jar has been copied
into lib/endorsed.
I decided to try an experiment using the jetty-j2ee Geronimo
assembly. I deleted the Xerces jars from the lib/endorsed
directory. I did this to convince myself that the java.endorsed.dirs
mechanism was working correctly and the problem was due to me missing
something with the yoko cofiguration. I expected Geronimo to "fall
over" during launch because of the missing jars. To my surprise, it
didn't.
I instrumented one of the yoko classes, and add it load the
IORInterceptor_3_0 class and org.w3c.dom.Element and dump the package
information for these classes. The Package information indicates
both of these classes are resolving to the JVM native versions rather
than the versions in lib/endorsed. The java.endorsed.dirs appears to
be set to the correct value, and the jar files are in the appropriate
directory, but the classes don't appear to be getting picked up.
This was just the last experiment for a problem I've been chasing
since Monday. I'm getting extremely inconsistent results from using
java.endorsed.dirs. Right now, I have 4 situations in front of me:
1) Simple standalone test case using Yoko ORB.
2) Yoko unit tests using surefire plugin.
3) Openejb Yoko unit tests using surefile plugin (setup was copied
from the Yoko unit tests).
4) Full Geronimo assembly using the Yoko ORB.
When running under Java 5, 1) and 2) work ok. 3) fails, even though
the surefile tests (in theory) are running exactly the same way as
2). 4) also fails. 1), 3), and 4) all work fine under Java 1.4.2,
but there's no conflict with the native IORInterceptor_3_0 class is
that scenario. I spent most of yesterday wrestling with scenario 3),
and never getting it to work. Like scenario 4), the jar files are in
the target endorsed directory and the java.endorsed.dirs is pointing
to the correct location. Right now, I'm completely stumped, and I'm
wondering if this has every really worked in the Geronimo assembly.
Rick
