Re: FYI: Updated "Building GCJ for Windows"

[Marco Trudel <mtrudel at gmx dot ch>]

Hello Ranjit

Thank you for the tutorial! Great work!
For GCC 4.1.1 I additionally needed these patches (didn't use the ones 
you mentioned):

Before building the compiler running on linux building for windows:
-patch-natSharedLibLoader 
(http://gcc.gnu.org/ml/java-patches/2006-q2/msg00091.html)
  Compilation of gcc will fail with "./java/lang/reflect/Modifier.h:41: 
error: expected unqualified-id before numeric constant" withhout this patch

Before building the compiler running on windows building for windows:
- patch-start (http://gcc.gnu.org/ml/java/2006-04/msg00010.html)
  Withhout this patch, compiled programs won't do anything on windows
- patch-jni (http://gcc.gnu.org/ml/java/2006-02/msg00066.html)
  Withhout this patch, JNI won't work (thus SWT fails)...


Open problems with gcc4.1.1 on windows (as far as I encountered):

- reflection (Class.forName(...)) is broken
Classes won't be found be Class.forName() (thus throwing an 
ClassNotFoundException).
Any Ideas? I once read something in the archives, but I'm really unable 
to find it again... I hope there's a solution for this.

- serialisation seems to be broken too
Writing to files works, loading again fails with (if I remember right) 
ClassNotFoundExceptions... Might be because of the reflection problem. 
my solution was to write good old ordinary config files

- stacktraces are unusuable
Whats the state? I'm relatively new to this list and it seems to be an 
old problem. Something like dwarf would be better than sjil exceptions 
but don't works on windows or would need a big deal to implement; 
there's no active windows maintainer so this keeps open? Am I more or 
less in the picture?

- uses of the regexp (e.g. String.split()) classes will fail with: 
java.util.MissingResourceException: Bundle gnu/regexp/MessagesBundle not 
found
Workaround: http://lists.debian.org/debian-gcc/2005/09/msg00141.html
better Workaround: put the gnu/regexp/MessagesBundle into a jar and 
compile it too, so it will be builtin.


regards
Marco


Ranjit Mathew wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi,
>
>   I've updated my "Building GCJ for Windows" article:
>
>   http://rmathew.com/articles/gcj/bldgcj.html
>
> as well as the associated scripts to work with the latest
> GCC sources.
>
> At the moment of this writing, you will need these patches:
>
>   http://gcc.gnu.org/ml/java-patches/2006-q2/msg00285.html
>   http://gcc.gnu.org/ml/gcc-patches/2006-06/msg00410.html
>
> for the mainline to be able to build GCJ for MinGW.
>
> MinGW really needs to get the libgcj.dll thing working.
> Right now each of the executables statically linked
> with libgcj (like gij.exe, jv-convert.exe, etc.) are *huge* - about
> 40MB before stripping off debugging information and about 15MB
> after that. :-(
>
> The compiler is able to create binaries and I've done almost
> no testing beyond "Hello World!", though I notice that the
> stack traces are completely busted.
>
> HTH,
> Ranjit.
>
> - --
> Ranjit Mathew       Email: rmathew AT gmail DOT com
>
> Bangalore, INDIA.     Web: http://rmathew.com/
>
>
>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.2 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iD8DBQFEiZ+hYb1hx2wRS48RAnAmAJ9s3ldi7zxjr4Im8t3KUk4F2L+7DgCcDA2g
> tRriUTZPMrpdqHJQ6QNhMQs=
> =SMMa
> -----END PGP SIGNATURE-----
--- gcc/unwind-sjlj.c
+++ gcc/unwind-sjlj.c
@@ -265,7 +265,7 @@
   else
     {
       fs->personality = context->fc->personality;
-      return _URC_NO_REASON;
+      return context->fc->prev != NULL ? _URC_NO_REASON : _URC_END_OF_STACK;
     }
 }

--- libjava/gnu/gcj/runtime/natSharedLibLoader.cc
+++ libjava/gnu/gcj/runtime/natSharedLibLoader.cc
@@ -11,11 +11,12 @@
 #include <config.h>
 
 // If we're using the Boehm GC, then we need this include to override dlopen.
+#include <gcj/cni.h>
+
 #ifdef HAVE_BOEHM_GC
 #include <gc.h>
 #endif /* HAVE_BOEHM_GC */
 
-#include <gcj/cni.h>
 #include <jvm.h>
 #include <execution.h>
 

--- libjava/configure.host
+++ libjava/configure.host
@@ -260,6 +260,7 @@
 	slow_pthread_self=
 	;;
   *-mingw*)
+	libgcj_flags="${libgcj_flags} -fno-omit-frame-pointer"
   	# FIXME: win32_exception_handler( ) in win32.cc does not do the
 	# right stuff yet w.r.t. SEH. Live with the following for now.
 	can_unwind_signal=no