quartz-wm 1.2.1_0 does not build on PPC Leopard because of "Undefined symbols"

[ballapete (Peter "Pete" Dyballa)]

A few times during the build process GCC reports:

frame.h:40:2: warning: #warning "Old libXplugin version detected.  Some features may not be supported."

In the end the build fails with:

  OBJCLD quartz-wm
Undefined symbols:
  "__dispatch_main_q", referenced from:
      __dispatch_main_q$non_lazy_ptr in dock-support.o
  "_dispatch_async", referenced from:
      _DockRestoreItemAsync in dock-support.o
      ___DockRestoreItemAsync_block_invoke_1 in dock-support.o
  "__NSConcreteStackBlock", referenced from:
      __NSConcreteStackBlock$non_lazy_ptr in dock-support.o
  "_dispatch_get_current_queue", referenced from:
      _dock_message_handler in dock-support.o
  "_dispatch_queue_create", referenced from:
      _DockInit in dock-support.o
ld: symbol(s) not found

There are also a lot of messages about "shadowed declarations".

[ballapete (Peter "Pete" Dyballa)]

main.log

[jeremyhu (Jeremy Huddleston Sequoia)]

Those should all be weak in dock-support.o

[ballapete (Peter "Pete" Dyballa)]

Replying to jeremyhu@…:

Those should all be weak in dock-support.o

If you mean with "weak" weak binding, then this happens *after* link time, at run time of the linked object.

When you mean with "weak" that the symbols do not need to be available at link time, then quartz-wm needs to be built using a two-level namespace.

[ballapete (Peter "Pete" Dyballa)]

Replying to jeremyhu@…:

Those should all be weak in dock-support.o

Maybe this excerpt from ld man page helps to solve the problem:

     -undefined treatment
                 Specifies how undefined symbols are to be treated. Options are: error, warning,
                 suppress, or dynamic_lookup.  The default is error.

[jeremyhu (Jeremy Huddleston Sequoia)]

What I mean is that these symbols are weak linked and checked for at runtime. The symbols are annotated with the weak_import attribute, so their absence shouldn't be an issue at link time.

[jeremyhu (Jeremy Huddleston Sequoia)]

Oh right, but they have to be there on the machine linking them... ugg...

[jeremyhu (Jeremy Huddleston Sequoia)]

Yeah, I guess the best solution really is -undefined suppress ... sigh

[jeremyhu (Jeremy Huddleston Sequoia)]

Actually, dynamic_lookup is probably better for the built on 10.5 running on 10.6 case

[jeremyhu (Jeremy Huddleston Sequoia)]

Can you please test this patch?

[jeremyhu (Jeremy Huddleston Sequoia)]

quartz-wm.patch

[ballapete (Peter "Pete" Dyballa)]

Replying to jeremyhu@…:

Can you please test this patch?

It looks much better now! Package is built. (I modified Portfile that the patch gets applied.)

[yaseppochi (Stephen J. Turnbull)]

Replying to jeremyhu@…:

Can you please test this patch?

The patch also allows "port upgrade quartz-wm" to succeed on Mac OS X 10.5.8 "Leopard" on a MacBook Pro Intel Core Duo machine. Before it failed with the same symptoms as above.

However, I had to edit the patch to remove the leading "a/" and "b/".

[jeremyhu (Jeremy Huddleston Sequoia)]

Thanks. r88178