It is very difficult to remove any lib shared library dependencies on solaris. While compiling the RPM sources on Solaris I wished to ensure my resulting executable was not dependent on there being a libz or berkeley DB installed on the system. One of my problems in compiling this was the linker options. (Other bug tickets are being opended with libtool and automake which caused other 'obvious' solutions to not work as well) The -static LDFLAG is interpred that the resulting binary MUST be static. It would be far more useful if the linkerdid its best to create a static binary. I wish to adjust this seach order for some of my compilations: > Specifically, on ELF and SunOS systems, `ld' will search a directory > for a library with an extension of `.so' before searching for one > with an extension of `.a'. Here is an example of my problems: 1) If I add '-static' to LDFLAGS I get an error: make LDFLAGS="-static" /bin/sh ../libtool --mode=link gcc -O2 -D_GNU_SOURCE -Wall -Wpointer-arith -Wstrict-prototypes -Wmissing-prototypes -Wno-char-subscripts -static -o dump dump.o ../build/librpmbuild.la ../lib/librpm.la ../rpmio/librpmio.la ../popt/libpopt.la mkdir .libs gcc -O2 -D_GNU_SOURCE -Wall -Wpointer-arith -Wstrict-prototypes -Wmissing-prototypes -Wno-char-subscripts -o dump dump.o ../build/.libs/librpmbuild.a ../lib/.libs/librpm.a -ldb-3.1 ../rpmio/.libs/librpmio.a -L/usr/local/lib -L/usr/ucblib -lnsl -lsocket -lz ../popt/.libs/libpopt.a -R/usr/local/lib -R/usr/ucblib -R/usr/local/lib -R/usr/ucblib Undefined first referenced symbol in file stpcpy ../rpmio/.libs/librpmio.a(rpmio.o) ld: fatal: Symbol referencing errors. No output written to dump 2) If I use make LDFLAGS="'-Wl,-Bstatic'" /bin/sh ../libtool --mode=link gcc -O2 -D_GNU_SOURCE -Wall -Wpointer-arith -Wstrict-prototypes -Wmissing-prototypes -Wno-char-subscripts '-Wl,-Bstatic' -o dump dump.o ../build/librpmbuild.la ../lib/librpm.la ../rpmio/librpmio.la ../popt/libpopt.la mkdir .libs gcc -O2 -D_GNU_SOURCE -Wall -Wpointer-arith -Wstrict-prototypes -Wmissing-prototypes -Wno-char-subscripts -Wl,-Bstatic -o .libs/dump dump.o ../build/.libs/librpmbuild.so ../lib/.libs/librpm.so -ldb-3.1 ../rpmio/.libs/librpmio.so -L/usr/local/lib -L/usr/ucblib -lnsl -lsocket -lz ../popt/.libs/libpopt.so -R/usr/local/lib -R/usr/ucblib ld: fatal: input of shared object `../build/.libs/librpmbuild.so' in static mode ld: fatal: input of shared object `../lib/.libs/librpm.so' in static mode ld: fatal: input of shared object `../rpmio/.libs/librpmio.so' in static mode ld: fatal: input of shared object `../popt/.libs/libpopt.so' in static mode ld: fatal: File processing errors. No output written to .libs/dump make[2]: *** [dump] Error 1
Is this the gcc linker on Solaris, or Sun's? Obviously, we can't do much about Sun's compiler. :)
I believe it is GNU linker. Definiatly GCC. To be honest it is really hard to tell what linker GCC is using or to set the linker properly when compiling. I have had major headaches compiling CGG so I may have a version which using the sun linker.
ld's -static affects all libraries and the resulting binary present on the command line, while -Bstatic and -Bdynamic is position dependent and you can just surround all occurences of -lz and -ldb* with -Wl,-Bstatic and -Wl,-Bdynamic. Alternatively, you can also replace -lz with full_path_to_your_libz_a/libz.a and similarly with the DB libraries, then that particular library will be linked in statically while the rest will be a dynamically linked application. I'm closing this because this is certainly not a bug in either binutils nor gcc.
I understand why you closed this ticket. Unfoturnatly there is some very poor interaction between libtool and ld because of the Solaris shared library model. Your suggestion just does not work in the context I need it to (see http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=16381 Additional comments from kestes.com 2000-08-17 12:17 ) The problem for me as a user is that both the libtool people and the ld people can easily complain that the bug is in the others tool. I opened a bug in both tools hoping someone would see that there was a reasonable fix in some tool. Now both tickets are closed and there are no fixes coming. Why is it so hard to get gcc to tell me what linker it uses?