jc1 segfaults, please help

[Adam Megacz <gcc at lists dot megacz dot com>]

Hey, is anybody here interested in figuring this out? I know squat
about the java frontend, so somebody more familiar with it can
probably figure out in a few minutes what would take me a few weeks.

This is the only thing preventing gcj-win32 from building "out of the
box". It's still functionally incomplete and buggy, but once this is
fixed, it should at least build all the way through and produce binaries
out of java code.

This is the problem, with stacktrace:

  http://gcc.gnu.org/ml/java/2002-01/msg00466.html

Here's an i686-linux-gnu-pc binary illustrating it:

  http://www.megacz.com/jc1

If you check out gcc right now and build with these configure/make
commands, you can see it crash for yourself:

  ../gcc/configure \
                 --target=i686-pc-mingw32 \
                 --host=i686-pc-linux-gnu \
                 --enable-languages=c,c++,java \
                 --disable-nls \
                 --with-gnu-ld \
                 --with-gnu-as \
                 --enable-libgcj \
                 --enable-gc-type=boehm \
                 --disable-shared \
                 --enable-threads=win32 \
                 --disable-hash-synchronization \
                 --disable-interpreter \
                 --enable-sjlj-exceptions

  make all-target-libjava

Here is Anthony Green's fix (which works):

  http://gcc.gnu.org/ml/gcc-patches/2002-02/msg00167.html

And Richard Henderson's spectacularly unhelpful comment "This is
wrong. Find out why":

  http://gcc.gnu.org/ml/gcc-patches/2002-02/msg00209.html

If it helps at all, the jc1 from CVS as of 12-Dec-2001 does not have
this flaw (12-Dec is where I branched my original mingw-gcj from).

  - a

-- 
"In the dream, someone was asking me 'So what you're saying is that
 all your dreams are lucid?' 'Yes,' I replied 'except for this one.'"
                                                         -- epepke@k5