On Mon, May 16, 2011 at 04:44:03PM +0200, Paolo Bonzini wrote: > On 05/16/2011 07:37 PM, Alon Levy wrote: > >I've settled on libtool > >as the linker solution since it seems to be the de facto standard, please > >correct me on this point if I'm wrong. We are not using autoconf, so > >I've called it directly in it's various modes, compile, link and install. > > I agree. I don't dislike this particularly :) if at all. > > >Legitemacy of the warning: basically libtool is concerned about linking in > >position > >dependant code. But the trace-dtrace.o contains no .rodata section and is > >not actually > >a problem for the linker, I've verfied the resulting library works fine with > >spicec (the > >spice client using it). > > I'm not sure, however, that glibc would still keep the sharing of .text. >
I'll check. > >I'd appreciate any suggestion of a replacement for libtool if it is not > >appropriate. > >Note that this building of a shared library is not something new - we've > >talked about > >this before in relation to libcacard inclusion inside qemu git repository. > > Do you really need tracing besides as a dependency of qemu_malloc? > Perhaps you can recompile qemu-malloc.o specially for libcacard, and > force usage of the nop backend. Sounds like a very good idea for short term. Otoh if I'm already building libcacard with qemu dependencies, it would be nice if I could benefit from the tracing framework. I'll investigate building trace-dtrace.lo and if this doesn't work I'll try nop backend. > > Paolo >
