[kaffe] Testing for 1.1.0 release

[mcmahill at mtl.mit.edu]

On Mon, 19 May 2003, Dalibor Topic wrote:

> --- mcmahill at mtl.mit.edu wrote:
> > 
> > 
> > anyone know what the latest status of alpha support is?  Ie, am I wasting
> > my time to even try on NetBSD/alpha?
> 
> I've created a patch that lets NetBSD/alpha progress much further than it did.
> It's attached. 
> 
> Now it fails to compile the class library using it's own java compiler due to
> an SIGFPE, among other things. This is how you can help:
> 
> * Get current rt.jar file from http://www.kaffe.org/~robilad/rt.jar
> * check out kaffe from CVS
> * apply the attached patch
> * ./configure --prefix=/tmp/mykaffe --enable-debug --with-engine=intrp
> --with-rt-jar=/path/to/rt.jar
> * make ; make install
> * cd libraries/javalib
> * KAFFE_DEBUG=gdb /tmp/mykaffe/bin/kaffe -cp kjc.jar at.dms.kjc.Main
> @essential.files
> * in gdb: run

running now...

> * and try to figure out why the SIGFPE happens when 4.9e-324 is compared
> against some other value. ;)

this usually happens when you do a floating point operation on an
unitialized variable.  like

foo() {
double x;

if(x >= 3.2)
....

the debugger will often show 'x' as having a value of something like
4.9e-324.

hmm.  looking at the gdb output:

0x160083ef4 in soft_dcmpl (v1=4.9406564584124654e-324, v2=inf) at
soft.c:673
673             else if (v1 > v2) {

it seems like there's an infinity too.  

I'll try to find a few minutes tonight to look for

1)  proper initialization of v1

2)  try compiling with -mieee

> or just run make check after make install and see if you can fix some of the
> (mostly floating point related) regression test failures ... I've got 10
> failures, which are attached. 4 of them are being worked on within the scope of
> class loader update to support java 1.2, so they are not netbsd or alpha
> specific.


DoubleComp.fail - probably wants IEEE-754 math.  To get this add -mieee to
CFLAGS.

LostFrame.fail  - which version of NetBSD are you on?  1.6 had some bugs
in ld.so_elf.  Should be fixed in 1.6.1


-Dan