GCJ/minGW produced executables and linux/wine

[Jeff Sturm]

On Tue, 4 Mar 2003, Ranjit Mathew wrote:
> > > "--enable-sjlj-exceptions" explicitly during GCC configuration, it
> > > builds as a DWARF-2 EH using target.
> >
> > Hmm... that's not what I observe.  Not on the 3.3 branch at least.
>
> Yes, once again, I confused the effects of mingw-local patches v/s
> vanilla GCC 3.3. Sorry.

Ugh.  How is it that mingw always has some outstanding patch set that
never quite gets integrated into GCC?  (Don't answer that, I don't
really want to know...)

I think this test isn't quite right:

if test $can_unwind_signal = no && test $enable_sjlj_exceptions = no; then
  CHECKREFSPEC=-fcheck-references
  ...

since it gives no assurances that signal handling a.k.a. exception filters
work at all.

I re-ran the testsuite with -fcheck-references.  It looks much better:


        === libjava Summary ===

# of expected passes        2297
# of unexpected failures    20
# of expected failures      16
# of untested testcases     27

If you agree that something like this should be in 3.3 (unless exception
handling is fixed), I'll prepare a patch.

Many of the remaining failures are caused by:

void
_Jv_ThreadInterrupt (_Jv_Thread_t *data)
{
  MessageBox(NULL, "Unimplemented", "win32-threads.cc:_Jv_ThreadInterrupt", MB_O  // FIXME:
}

which also causes my testsuite to hang waiting for mouse clicks :-(

If these and gctest can get fixed somehow, I'd say mingw-gcj looks good
for the 3.3 release (what is the timeframe, again?)  Otherwise,
status.html could be amended to say something more helpful than "Is
incomplete".

Jeff