--whole-archive doesn't work on OSX

[Andrew Miller]

Hi,

I have just posted a bug regarding libtool on OSX to 
http://savannah.gnu.org/support/index.php?func=detailitem&item_id=105489

I'm not sure if that is the right place to post bugs, so in case it 
isn't, here is the text of the report:

This bug tested on:
Darwin bioeng21.bioeng.auckland.ac.nz 8.7.0 Darwin Kernel Version 8.7.0: 
Fri May 26 15:20:53 PDT 2006; root:xnu-792.6.76.obj~1/RELEASE_PPC Power 
Macintosh powerpc

It seems that recent versions of libtool don't support the 
--whole-archive on MacOSX. This causes the --whole-archive flag to be 
passed to gcc (which seems to be silently ignored on OSX), resulting in 
a link which doesn't really contain the whole archive.
For example:

/bin/sh ./libtool --tag=CXX --mode=link g++ -g -O2 -o 
cellml_corba_server -lxml2 -lomniORB4 -lomnithread --whole-archive 
cellml_corba_server-Main.o ./libcellml_context.la 
./libcellml_context_corba.la ./libcellml.la ./libcellml_corba.la 
./libCORBASupport.la ./libltdlc.la
g++ -g -O2 -o .libs/cellml_corba_server --whole-archive 
cellml_corba_server-Main.o -Wl,-bind_at_load /usr/lib/libxml2.dylib 
-lpthread -lz /usr/lib/libiconv.dylib -lm -lomniORB4 -lomnithread 
./.libs/libcellml_context.dylib ./.libs/libcellml_context_corba.dylib 
./.libs/libcellml.dylib ./.libs/libcellml_corba.dylib 
./.libs/libCORBASupport.a ./.libs/libltdlc.a
creating cellml_corba_server

bioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$ nm 
.libs/cellml_corba_server |grep gWrapperRepo
bioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$ nm 
.libs/libCORBASupport.a |grep gWrapperRepo
000000ac t __GLOBAL__D__Z18gWrapperRepositoryv
00000000 a __GLOBAL__D__Z18gWrapperRepositoryv.eh
0000009c t __GLOBAL__I__Z18gWrapperRepositoryv
00000000 a __GLOBAL__I__Z18gWrapperRepositoryv.eh
00000190 T __Z18gWrapperRepositoryv
00001820 S __Z18gWrapperRepositoryv.eh
00001a58 b __ZGVZ18gWrapperRepositoryvE2wr
00001a60 b __ZZ18gWrapperRepositoryvE2wr

I understand older libtools gave the -Wl,-all_load flag, which is 
similar to --whole-archive on the ld from GNU binutils. However, unlike 
GNU ld, you can't turn off all_load for some archives, and leave it on 
for others, so you end up messing up the linking of the implicit libraries:
^ioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$ g++ 
-Wl,-all_load -g -O2 -o .libs/cellml_corba_server --whole-archive 
cellml_corba_server-Main.o -Wl,-bind_at_load /usr/lib/libxml2.dylib 
-lpthread -lz /usr/lib/libiconv.dylib -lm -lomniORB4 -lomnithread 
./.libs/libcellml_context.dylib ./.libs/libcellml_context_corba.dylib 
./.libs/libcellml.dylib ./.libs/libcellml_corba.dylib 
./.libs/libCORBASupport.a ./.libs/libltdlc.a
/usr/bin/ld: warning multiple definitions of symbol ___eprintf
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc.a(_eprintf.o) private 
external definition of ___eprintf in section (__TEXT,__text)
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libstdc++.dylib(single module) 
definition of ___eprintf
/usr/bin/ld: warning multiple definitions of symbol ___divdi3
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc.a(_divdi3.o) private 
external definition of ___divdi3 in section (__TEXT,__text)
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc_s.dylib(_divdi3_s.o) 
definition of ___divdi3
/usr/bin/ld: warning multiple definitions of symbol ___udivdi3
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc.a(_udivdi3.o) private 
external definition of ___udivdi3 in section (__TEXT,__text)
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc_s.dylib(_udivdi3_s.o) 
definition of ___udivdi3
/usr/bin/ld: warning multiple definitions of symbol ___umoddi3
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc.a(_umoddi3.o) private 
external definition of ___umoddi3 in section (__TEXT,__text)
/usr/lib/gcc/powerpc-apple-darwin8/4.0.0/libgcc_s.dylib(_umoddi3_s.o) 
definition of ___umoddi3

(which is presumably why older versions dropped that flag, and caused 
this bug).

Unless someone finds a better way, I think the correct solution will 
have to be to unarchive the convenience libraries and link the objects 
manually.

Best regards,
Andrew Miller