harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexei Fedotov" <alexei.fedo...@gmail.com>
Subject Re: [eclipse] Eclipse 3.3 crashes on current Harmony
Date Thu, 06 Sep 2007 05:16:08 GMT
Alexey,

Sorry. At the time I wrote that observation I was confused with the
alternative verifier convention to return _FALSE to indicate a
successful check in checkAccess* functions.

The problem happens in a different place and is related to mixing of
classes from different class loaders. The fact is that compared class
handles reference different class loaders, and this might be problem
of the verifier internal cache.

I've updated HARMONY-4738 [1] with the latest information, but
resolution is not clear yet.

Thanks.

[1] http://issues.apache.org/jira/browse/HARMONY-4738


On 9/6/07, Alexey Petrenko <alexey.a.petrenko@gmail.com> wrote:
> Thanks, Alexei.
>
> So I would say that we should revert this patch and then think on how
> to resolve the issue in better way?
>
> Thoughts?
>
> SY, Alexey
>
> 2007/9/5, Alexei Fedotov <alexei.fedotov@gmail.com>:
> > Gregory,
> > Thanks for your help!
> >
> > My observation is that the problem showed itself after the following fix
> > http://svn.apache.org/viewvc?view=rev&revision=563763
> >
> > Tracking things in a debugger I can see that the following new line of
> > code returns unexpected result. I will dig further into the behavior
> > of this function.
> > -        return _TRUE;
> > +        return !class_is_same_package(k_class, referred) &&
> > vf_is_extending(k_class, referred);
> >
> > With best regards, Alexei
> >
> > On 9/5/07, Mark Hindess <mark.hindess@googlemail.com> wrote:
> > >
> > > On 5 September 2007 at 16:05, Gregory Shimansky <gshimansky@gmail.com>
wrote:
> > > > Alexey Petrenko wrote:
> > > > > 2007/9/4, Gregory Shimansky <gshimansky@gmail.com>:
> > > > >> Gregory Shimansky wrote:
> > > > >>> It seems to be a bug in verifier, but there weren't any changes
in
> > > > >>> verifier recently, so I am not sure it is a regression. Probably
this
> > > > >>> bug was hidden and didn't show up until recently.
> > > > >> BTW, building DRLVM with -Duse_original_verifier=true allows
eclipse 3.3
> > > > >> to start. I think that maybe switching back to older verifier
again
> > > > >> might improve the situation.
> > > > > Yes, this helps. -noverify should also help in this case.
> > > > >
> > > > > I've created HARMONY-4738 to track the issue.
> > > > > It would be nice if someone familiar with the verifier could fix
the issue.
> > > >
> > > > BTW I think I've found out the reason for this bug. Eclipse 3.3 has an
> > > > ICU4J implementation bundled with it. It uses ICU 3.6.1 while we have
> > > > ICU4J 3.4.4 in bootclasspath.
> > > >
> > > > In this case ICU classes are loaded from our 3.4.4 JAR file except for
> > > > some classes that may not be present in the older version. In this case
> > > > they are taken from Eclipse's JAR file which leads to inconsistencies.
> > > >
> > > > I wonder why older verifier implementation works ok...
> > >
> > > No idea.  I also wonder why J9 is unaffected.
> > >
> > > FYI: I see similar but not identical problems on Linux.
> > >
> > > -Mark.
> > >
> > > > > https://issues.apache.org/jira/browse/HARMONY-4738
> > > > >
> > > > >>>> 2007/9/4, Alexey Petrenko <alexey.a.petrenko@gmail.com>:
> > > > >>>>> Yes, looks like we really have a regression.
> > > > >>>>> I've tried Eclipse with M2 and recent classlib with
j9 and works in
> > > > >>>>> both cases. But silently crashes silently in the
very beging. Even
> > > > >>>>> without trace in configuration directory.
> > > > >>>>>
> > > > >>>>> So we need to investigate the issue.
> > > > >>>>>
> > > > >>>>> Any help is welcome :)
> > > > >>>>>
> > > > >>>>> SY, Alexey
> > > > >>>>>
> > > > >>>>> 2007/9/4, Tim Ellison <t.p.ellison@gmail.com>:
> > > > >>>>>> Leo Li wrote:
> > > > >>>>>>> On 9/4/07, Alexey Petrenko <alexey.a.petrenko@gmail.com>
wrote:
> > > > >>>>>>>> Guys,
> > > > >>>>>>>>
> > > > >>>>>>>> I've tried to run In Harmony with Eclipse
bundle on the recent
> > > > >>>>>>>> harmony/drlvm build but it crashes silently
in the very begining.
> > > > >>>>>>>> I would say that this is very serious
regression for us.
> > > > >>>>>>>>
> > > > >>>>>>>> I'll continue investigation but it would
be nice if someone will try
> > > > >>>>>>>> to investigate it in the same time.
> > > > >>>>>>>>
> > > > >>>>>>>> Thanks in advance.
> > > > >>>>>>>>
> > > > >>>>>>>> SY, Alexey
> > > > >>>>>>>>
> > > > >>>>>>>  Yes, it is a serious regression. And do
you know when it start to
> > > > >>>>>>> fail?
> > > > >>>>>>>   I will try it when I go to office tomorrow
since I have no classlib
> > > > >>>>>>> on hand. Not sure whether I can help.:)
> > > > >>>>>> FYI: It's working ok for me on the current HEAD
of classlib (r572680)
> > > > >>>>>> with the IBM VME.
> > > > >>>>>>
> > > > >>>>>> Regards,
> > > > >>>>>> Tim
> > > > >>>>>>
> > > > >>>
> > > > >>
> > > > >> --
> > > > >> Gregory
> > > > >>
> > > > >>
> > > > >
> > > >
> > > >
> > > > --
> > > > Gregory
> > > >
> > >
> > >
> > >
> >
> >
> > --
> > With best regards,
> > Alexei,
> > ESSD, Intel
> >
>


-- 
With best regards,
Alexei,
ESSD, Intel

Mime
View raw message