darwin & libgcj ?

darwin & libgcj ?

[Andreas Tobler]

Can we enable libgcj on powerpc-darwin by default?

It builds with no problems out of the box with --enable-libgcj on both 
3.3 and trunk.

Any objections?

Thanks,

Andreas

darwin & libgcj ?

[Andrew Haley]

Andreas Tobler writes:
 > Can we enable libgcj on powerpc-darwin by default?
 > 
 > It builds with no problems out of the box with --enable-libgcj on both 
 > 3.3 and trunk.
 > 
 > Any objections?

Testsuite results would be nice.  I don't want it built if it doesn't
work.

Andrew.

Re: darwin & libgcj ?

[Andreas Tobler]

Andrew Haley wrote:
> Andreas Tobler writes:
>  > Can we enable libgcj on powerpc-darwin by default?
>  >  > It builds with no problems out of the box with --enable-libgcj on both 
>  > 3.3 and trunk.
>  >  > Any objections?
> 
> Testsuite results would be nice.  I don't want it built if it doesn't
> work.

Well, I agree here. I build it since a longer time. But unfortunately 
the results look pretty bad due to a multiple definition of read. Expect 
gets confused and dropps the test:

Running 
/Volumes/xufs/gcc-branch/gcc/libjava/testsuite/libjava.lang/lang.exp ...
Executing on host: 
/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3/libj
ava/testsuite/../libtool --silent --tag=GCJ --mode=link 
/Volumes/reserved1/gcchf
s/objdir/gcc/gcj -B/Volumes/reserved1/gcchfs/objdir/gcc/ 
--encoding=UTF-8 -B/Vol
umes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjava/ 
/Volumes/xufs/gcc
-branch/gcc/libjava/testsuite/libjava.lang/ArrayStore.java   -no-install 
--main=
ArrayStore -g 
-L/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjav
a/.libs 
-L/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//boehm-gc/.li
bs    -o 
/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3/libjava/testsu
ite/ArrayStore.exe    (timeout = 300)
/usr/bin/ld: warning multiple definitions of symbol _read^M
/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjava/.libs/libgcj.a
(os_dep.o) definition of _read in section (__TEXT,__text)^M
/usr/lib/libm.dylib(read.So) definition of _read^M
output is:
/usr/bin/ld: warning multiple definitions of symbol _read^M
/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjava/.libs/libgcj.a
(os_dep.o) definition of _read in section (__TEXT,__text)^M
/usr/lib/libm.dylib(read.So) definition of _read^M


FAIL: ArrayStore compilation from source
UNTESTED: ArrayStore execution from source compiled test
UNTESTED: ArrayStore output from source compiled test
PASS: ArrayStore byte compilation
ERROR: No `main' given in program ArrayStore

When I run the test by 'hand' it works fine.

Andreas


                 === libjava Summary ===

# of expected passes            484
# of unexpected failures        238
# of unresolved testcases       159
# of untested testcases         751
runtest completed at Sat Jan  4 10:01:29 2003

Re: darwin & libgcj ?

[Andrew Haley]

Andreas Tobler writes:
 > Andrew Haley wrote:
 > > Andreas Tobler writes:
 > >  > Can we enable libgcj on powerpc-darwin by default?
 > >  >  > It builds with no problems out of the box with --enable-libgcj on both 
 > >  > 3.3 and trunk.
 > >  >  > Any objections?
 > > 
 > > Testsuite results would be nice.  I don't want it built if it doesn't
 > > work.
 > 
 > Well, I agree here. I build it since a longer time. But unfortunately 
 > the results look pretty bad due to a multiple definition of read. Expect 
 > gets confused and dropps the test:

Is this some weirdness in the darwin implementation of DSOs?  We
redefine all sorts of things in libgcj, not just that one.

Andrew.

Re: darwin & libgcj ?

[Andreas Tobler]

Andrew Haley wrote:
> Andreas Tobler writes:
>  > Andrew Haley wrote:
>  > > Andreas Tobler writes:
>  > >  > Can we enable libgcj on powerpc-darwin by default?
>  > >  >  > It builds with no problems out of the box with --enable-libgcj on both 
>  > >  > 3.3 and trunk.
>  > >  >  > Any objections?
>  > >  > > Testsuite results would be nice.  I don't want it built if it doesn't
>  > > work.
>  >  > Well, I agree here. I build it since a longer time. But unfortunately 
>  > the results look pretty bad due to a multiple definition of read. Expect 
>  > gets confused and dropps the test:
> 
> Is this some weirdness in the darwin implementation of DSOs?  We
> redefine all sorts of things in libgcj, not just that one.

I never cared about since the apps were running. Probably we should ask 
some Apple people?

Andreas

Re: darwin & libgcj ?

[Andreas Tobler]

Andrew Haley wrote:

> Is this some weirdness in the darwin implementation of DSOs?  We
> redefine all sorts of things in libgcj, not just that one.

/Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjava/.libs/libgcj.a
(os_dep.o) definition of _read in section (__TEXT,__text)^M
/usr/lib/libm.dylib(read.So) definition of _read^M

Btw, this multiple read is coming from boehm-gc.
Though I don't know if it is correct.
Andreas

Re: darwin & libgcj ?

[Geoff Keating]

Andrew Haley <aph@redhat.com> writes:

> Andreas Tobler writes:
>  > Andrew Haley wrote:
>  > > Andreas Tobler writes:
>  > >  > Can we enable libgcj on powerpc-darwin by default?
>  > >  >  > It builds with no problems out of the box with --enable-libgcj on both 
>  > >  > 3.3 and trunk.
>  > >  >  > Any objections?
>  > > 
>  > > Testsuite results would be nice.  I don't want it built if it doesn't
>  > > work.
>  > 
>  > Well, I agree here. I build it since a longer time. But unfortunately 
>  > the results look pretty bad due to a multiple definition of read. Expect 
>  > gets confused and dropps the test:
> 
> Is this some weirdness in the darwin implementation of DSOs?  We
> redefine all sorts of things in libgcj, not just that one.

It's a warning, that can be switched off with the -multiply_defined
(defaults to "warning") and -multiply_defined_unused (defaults to
"suppress") flags.  Probably the other redefinitions are never used by
that test program so the message doesn't appear.

This probably isn't doing what the code hopes.  The read() defined in
another shared library or the application won't be called by system
libraries, and I'm not sure that Darwin even makes any promises about
the ordering of shared libraries, so it's possible that the read()
doesn't even get called by applications.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: darwin & libgcj ?

[Andrew Haley]

Andreas Tobler writes:
 > Andrew Haley wrote:
 > 
 > > Is this some weirdness in the darwin implementation of DSOs?  We
 > > redefine all sorts of things in libgcj, not just that one.
 > 
 > /Volumes/reserved1/gcchfs/objdir/powerpc-apple-darwin6.3//libjava/.libs/libgcj.a
 > (os_dep.o) definition of _read in section (__TEXT,__text)^M
 > /usr/lib/libm.dylib(read.So) definition of _read^M
 > 
 > Btw, this multiple read is coming from boehm-gc.

Aha!  Well, that is wrong.  Perhaps the linker is failing to do the
right thing.  But without a system to debug I don't know what is
causing the problem.

Andrew.

Re: darwin & libgcj ?

[Jeff Sturm]

On Sun, 5 Jan 2003, Andrew Haley wrote:
>  > Btw, this multiple read is coming from boehm-gc.
>
> Aha!  Well, that is wrong.  Perhaps the linker is failing to do the
> right thing.  But without a system to debug I don't know what is
> causing the problem.

I think the simplest thing to do is disable incremental GC on Darwin; see
patch below.  (AFAICT it didn't work anyway; gctest seemed to hang
forever.)

With this small patch I get some success on Darwin:

                === libjava Summary ===

# of expected passes            335
# of unexpected failures        22
# of expected failures          2
# of unresolved testcases       91
# of untested testcases         16


Index: include/private/gcconfig.h
===================================================================
RCS file: /cvs/gcc/gcc/boehm-gc/include/private/gcconfig.h,v
retrieving revision 1.26
diff -u -5 -p -r1.26 gcconfig.h
--- include/private/gcconfig.h  2 Oct 2002 06:25:37 -0000       1.26
+++ include/private/gcconfig.h  6 Jan 2003 01:05:40 -0000
@@ -683,11 +683,11 @@
 #     define ALIGNMENT 4
 #     define OS_TYPE "MACOSX"
 #     define DATASTART ((ptr_t) get_etext())
 #     define STACKBOTTOM ((ptr_t) 0xc0000000)
 #     define DATAEND   /* not needed */
-#     define MPROTECT_VDB
+#     undef MPROTECT_VDB
 #     include <unistd.h>
 #     define GETPAGESIZE() getpagesize()
 #   endif
 #   ifdef NETBSD
 #     define ALIGNMENT 4