harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark Hindess <mark.hind...@googlemail.com>
Subject Re: Next milestones: 5.0M16 and 6.0M4
Date Thu, 09 Dec 2010 10:11:05 GMT

In message <AANLkTi=+qfDQUO2m7nw8+ZWgLba986k7p-sxYaR-dJ90@mail.gmail.com>,
Ray Chen writes:
>
> On Mon, Nov 29, 2010 at 10:08 PM, Mark Hindess
> <mark.hindess@googlemail.com> wrote:
> >
> > Your enthusiasm is commendable but if we still don't have a
> > volunteer to be release manager then there wont be any milestone
> > releases.
> >
> > When I integrated my build changes I made a mistake and subsequently
> > jdwp was being built in release mode when it hadn't been previously.
> > The optimized code exhibits a problem - that I guess must exist in
> > some of our previous milestone releases - that isn't seen in debug
> > mode.  I think this is causing the JDWP failures.
> >
> > I had planned to binary chop the sources to see which one has
> > problems when optimized but haven't had time to do this yet.  For
> > now I guess we should just disable optimization for the jdwp module.
>
> Can you point it out how to disable optimization for the jdwp
> module? Shall we modify some scripts?

So, it turns out I was wrong about where the problem is...  I did two
builds:

  ant -Dhy.cfg=release -Dhy.target=$PWD/target.release

and:

  ant -Dhy.cfg=debug -Dhy.target=$PWD/target.debug

I then ran the org.apache.harmony.jpda.tests.jdwp.Method.VariableTableTest
on each one.  It fails on the release build and works on the debug build.

So I created a test build by initially copying the working target.debug:

  cp -pr target.debug target

Then I did:

  cp target.release/hdk/jdk/jre/bin/libjdwp.so target/hdk/jdk/jre/bin
  cp target.release/hdk/jdk/jre/bin/libdt_socket.so target/hdk/jdk/jre/bin

expecting this to cause it to stop working but the test still passed.  In
fact if I replaced the entire working (debug) jre directory with the jre
directory from the failing (release) one, then it still passes.

It turns out that hdk/build/test/jpda/jpda_tests.jar is the crucial file
that causes it to break or not.  To get further I need to binary chop
this file to find out what is different between the working (debug) copy
of this jar and the failing (release) copy.

I've not got time to do this - at least not before the code freeze
tomorrow.  Also, I'd be interested to know if anyone else can reproduce
my results as I've been doing this in my non-existent spare time over
the course of about three weeks so it is quite possible I made mistakes.

I will carry on looking at it as/when I get spare time.

Regards,
 Mark.





Mime
View raw message