On 7/22/2013 3:37 PM, Jochen Theodorou wrote:
Am 20.07.2013 03:41, schrieb Mandy Chung:
Hi Jochen,
I read through the thread in mlvm-dev [1] that has a good discussion
there. I have filed a RFE:
8020785: caller-sensitive methods to skip dynamic generated frames
and look up the true caller
This seems that java.lang.instrument might be an appropriate place for
this support. This certainly requires further investigation.
Mandy
[1]
http://mail.openjdk.java.net/pipermail/mlvm-dev/2013-July/005387.html
[2] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=8020785
can you explain a little how instrumentation could help here? Of
course I don't expect a full solution, but the idea would be helping.
As I understand your use case, what you need is to tell the system to
skip some specific frames when a caller-sensitive method is looking up
the caller which the functionality depends on (e.g.
ResourceBundle.getBundle and Class.forName uses the caller's loader to
find class/resource). You currently achieve this by walking the stack.
I like Charles's suggestion to teach JVM about language-specific stack
traces. Instrumentation agents probably do something similar to skip
its own runtime classes / packages.
The current hotspot implementation skips sun.reflect.* and a few other
specific classes to determine the caller class. I think
java.lang.instrument seems an appropriate place for this kind of
functionality and provide a way to extend the whitelist to skip and pass
it to the JVM. This is really just a rough idea and there may be issues
to be investigated.
I didn't have time to look into this RFE as I have been very busy on
other high priority works. Whoever works on this RFE will certainly
discuss the proposals in core-libs-dev and get feedbacks.
Mandy
