Peter O'Gorman wrote:
On Aug 3, 2006, at 10:01 AM, Andrew Miller wrote:
...
However, the documentation seems to imply that the behaviour of
whole-archive should happen on all platforms, for all convenience
libraries.
e.g.
— Variable: *whole_archive_flag_spec*
Compiler flag to generate shared objects from convenience archives.
If whole_archive_flag_spec is empty, as it is on darwin for more
recent versions of libtool then the convenience archive will be
unpacked using lipo/ar and the objects will be added to the link line.
I haven't been able to reproduce this behaviour when building binaries
for either libtool-1.5.22 or current CVS libtool (although it does
unpack the archive when building installable libraries from convenience
libraries).
Here is the output I am getting from a current CVS libtool...
/bin/sh ./libtool --tag=CXX --mode=link g++ -g -O2 -o
cellml_corba_server -lxml2 -lomniORB4 -lomnithread
cellml_corba_server-Main.o ./libcellml_context.la
./libcellml_context_corba.la ./libcellml.la ./libcellml_corba.la
./libCORBASupport.la ./libltdlc.la
libtool: link: g++ -g -O2 -o .libs/cellml_corba_server
cellml_corba_server-Main.o -Wl,-bind_at_load
/Users/cmiss/physiome_builds/trees/CellML_DOM_API_OSX/./.libs/dlopen.a
/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
This produces a cellml_corba_server binary which is missing symbols
which are in the convenience libraries:
bioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$ nm
./.libs/cellml_corba_server |grep __Z18gWrapperRepositoryv
bioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$ nm
./.libs/libCORBASupport.a |grep __Z18gWrapperRepositoryv
00ac t __GLOBAL__D__Z18gWrapperRepositoryv
a __GLOBAL__D__Z18gWrapperRepositoryv.eh
009c t __GLOBAL__I__Z18gWrapperRepositoryv
a __GLOBAL__I__Z18gWrapperRepositoryv.eh
0190 T __Z18gWrapperRepositoryv
1820 S __Z18gWrapperRepositoryv.eh
and when a module (which references __Z18gWrapperRepositoryv) is
dynamically loaded into cellml_corba_server, it results in:
bioeng21:~/physiome_builds/trees/CellML_DOM_API_OSX cmiss$
./cellml_corba_server
dyld: lazy symbol binding failed: Symbol not found: __Z18gWrapperRepositoryv
Referenced from:
/Users/cmiss/physiome_builds/trees/CellML_DOM_API_OSX/.libs/libcellml_context_corba.0.dylib
Expected in: flat namespace
dyld: Symbol not found: __Z18gWrapperRepositoryv
Referenced from:
/Users/cmiss/physiome_builds/trees/CellML_DOM_API_OSX/.libs/libcellml_context_corba.0.dylib
Expected in: flat namespace
Trace/BPT trap
This works correctly on Linux (it doesn't actually need
-Wl,--whole-archive here, because it gives the .a archives by name
rather than with -l, meaning that the whole archive gets linked anyway):
/bin/sh ./libtool --tag=CXX --mode=link g++ -g -O2 -o
cellml_corba_server -lxml2 -lomniORB4 -lomnithread
cellml_corba_server-Main.o ./libcellml_context.la
./libcellml_context_corba.la ./libcellml.la ./libcellml_corba.la
./libCORBASupport.la ./libltdlc.la -ldl
libtool: link: g++ -g -O2 -o .libs/cellml_corba_server
cellml_corba_server-Main.o
/people/amil082/code/CellML_DOM_API/./.libs/dlopen.a
/home/amil082/usr//lib/libxml2.so -lz -lm -lomniORB4 -lomnithread
./.libs/libcellml_context.so ./.libs/libcellml_context_corba.so
./.libs/libcellml.so ./.libs/libcellml_corba.so
./.libs/libCORBASupport.a ./.libs/libltdlc.a -ldl -Wl,-rpath
-Wl,/home/amil082/usr//lib -Wl,-rpath -Wl,/home/amil082/tmp/cda/lib
[EMAIL PROTECTED] CellML_DOM_API]$ nm ./.libs/cellml_corba_server |grep
gWrapperRepo
0804af20 t _GLOBAL__D__Z18gWrapperRepositoryv
0804af30 t _GLOBAL__I__Z18gWrapperRepositoryv
0804af50 T _Z18gWrapperRepositoryv
080526b0 b _ZGVZ18gWrapperRepositoryvE2wr
080526b8 b _ZZ18gWrapperRepositoryvE2wr
[EMAIL PROTECTED] CellML_DOM_API]$ ./cellml_corba_server
Extracting the objects from the archives for binary builds as well as
library builds for platforms where whole-archive is empty should fix this.
Best regards,
Andrew Miller
___
Bug-libtool mailing list
Bug-libtool@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-libtool
