RE: _Jv_CallAnyMehtodA alignment patch

["Boehm, Hans" <hans_boehm at hp dot com>]

I've attached a patch that expands an existing jni test to demonstrate the problem.  Unfortunately, in many cases this won't actually fail without special measures, since the hardware and/or kernel patch up misaligned references by default.  (On IA64 you need the prctl command to get it to fail properly, though a warning message will appear by default.)

Are both the test and the original patch (http://gcc.gnu.org/ml/java-patches/2003-q1/msg00317.html) OK for the trunk?

I would be inclined to also check these into 3.3, since unaligned references are clearly bugs, even if kernels and hardware sometimes try to hide them.  Other opinions?

Hans

> -----Original Message-----
> From: Tom Tromey [mailto:tromey at redhat dot com]
> Sent: Monday, February 10, 2003 3:51 PM
> To: Boehm, Hans
> Cc: 'Andrew Haley'; 'java-patches at gcc dot gnu dot org '
> Subject: Re: _Jv_CallAnyMehtodA alignment patch
> 
> 
> >>>>> "Hans" == Boehm, Hans <hans_boehm at hp dot com> writes:
> 
> Hans> (Unfortunately, it takes a few awt patches to get to this stage.
> Hans> I would like to see those, or corrected versions of them, make
> Hans> it into the tree as well.)
> 
> No problem.  I'm glad you've looked at this.
> 
> Hans> I think this should show up if you try to pass a byte as the
> Hans> first argument and a long as the second argument to a JNI
> Hans> method.  It looks like there are no such cases in the test
> Hans> suite.  The gtk peers seem to be by far the most demanding JNI
> Hans> test we have?
> 
> That's definitely true.  The tests in testsuite/libjava.jni/ are
> pretty minimal.
> 
> Could you write a new one?  If not, and if I write a new one, could
> you test it?
> 
> Tom
>

Attachment: 0219tests.diff
Description: Binary data