Re: [cp-patches] [patch] make darwin 'work' with classpath

[Andreas Tobler]

Hi Mark,

Mark Wielaard wrote:

> On Tue, 2005-08-02 at 06:41 +0200, Andreas Tobler wrote:
>
> > Well, here I should have said 'once', with the actual cp I didn't. :(
> > But I know jamvm will fail because it in src/dll.c:300 it checks against 
> > .so and new will .dylib on darwin.

In the meantime I built cp with .dylib and with .so. Also jamvm was 
tuned to work with both. (Not simultaneously, but this should be a 
little effort)
Unfortunately cp cvs does crash when I run a swing app. An awt based app 
does somehow work with jamvm, but only when I build cp with .dylib. 
Built with .so it segfaults. For both, awt and swing based apps.

The awt based app seems to work when I resize the frame (I tried 
awt.Demo). If I do not resize, I get a blank frame with the menubar 
only. Resizing the frame brings me the buttons to see, except the close 
button. This is all against gnome-2.10 core.

The swing based app aborts with the following message:

** ERROR **: file gnu_java_awt_peer_gtk_GdkGraphics.c: line 161 
(Java_gnu_java_awt_peer_gtk_GdkGraphics_initFromImage): assertion 
failed: (pixmap != NULL)
aborting...
java.lang.InternalError
   at gnu.java.awt.peer.gtk.GdkGraphics.initFromImage (Native Method)
   at gnu.java.awt.peer.gtk.GdkGraphics.<init> (GdkGraphics.java:107)
   at gnu.java.awt.peer.gtk.GtkImage.getGraphics (GtkImage.java:311)
   at gnu.java.awt.peer.gtk.GtkComponentPeer.createImage 
(GtkComponentPeer.java:234)
   at java.awt.Component.createImage (Component.java:2058)
   at java.awt.Component.createImage (Component.java:2056)
   at javax.swing.RepaintManager.getOffscreenBuffer 
(RepaintManager.java:487)
   at javax.swing.JComponent.paint (JComponent.java:1440)
   at java.awt.Container$GfxPaintVisitor.visit (Container.java:1737)
   at java.awt.Container.visitChild (Container.java:1541)
   at java.awt.Container.visitChildren (Container.java:1503)
   at java.awt.Container.paint (Container.java:752)
   at gnu.java.awt.peer.gtk.GtkComponentPeer.handleEvent 
(GtkComponentPeer.java:311)
   at java.awt.Component.dispatchEvent (Component.java:2320)
   at java.awt.EventQueue.dispatchEvent (EventQueue.java:518)
   at java.awt.EventDispatchThread.run (EventDispatchThread.java:75)

Program received signal SIGABRT, Aborted.
[Switching to process 2154 local thread 0x3407]
0x9002ca98 in semaphore_wait_signal_trap ()

Ok, maybe it's a general ppc problem?

> OK. Useful additional datapoint :)
> Maybe the simplest solution then would be to symlink the .dylib to a .so
> with the same name as an additional workaround for darwin based systems.

No, .so and .dylib are not the same and you can not symlink apples to 
pears :)

> I looked a bit around and also found the extension .jnilib used on
> darwin systems for libraries loaded through jni. There is a patch for
> libtool to add a -jnimodule option, but I don't believe that has been
> integrated.
> http://lists.gnu.org/archive/html/libtool/2002-07/msg00010.html


You can dlopen/link against dylib. See the testsuite in 
gcc/libjava/testsuite/jni.exp there we test against. So I think we do 
not need an additional extension.

> Chris had a fight around this same issue for our libxmlj:
> http://lists.gnu.org/archive/html/classpath/2004-07/msg00207.html

But this was something different, he ended up using a wrong jni.h.

> It seems it is all a bit of a maze of extensions and options I am
> afraid :(

I know. But I can't change it :)

Regards,
Andreas