Axel asked: > First of all, thanks for finding out about it. But then, why do we > not > want [1]? Why is -nostdlib used for a library that should be used in > BeOS R5? Because in default libs there'ss R5 libnet.so. And linking a replacment libnet.so against R5 one don't sounds a good idea, does it!? More specificly, under vanilla R5, Be's libnet.so export a _res data symbol with a different size than ours (mostly because since ISC modified the _res_state struct concent), and gcc linker in this case, by default, choose the biggest, which is Be's. But in fact it don't, and make our _res symbol imported from Be's libnet.so. So, we make our libnet.so (and only this library) build with -nostdlib to explicitly tell the linker to link against only libroot.so. But, I overlooked at gcc link options, and choose the incorrect flag, removing startfiles at same time. > > Anyway, so I got curious about these magic files. So I went > > looking. > > Just look at our repository under src/kernel/glue and find all of > them > there :) > As you can easily see, init_term_dyn.o must be part of the object to > have the initialize_before/after() mechanism working. Which changing the -nostdlib flag to -nodefaultlibs in libnet.so Jamfile fix, as starfiles are linked again now. - Philippe -- Fortune Cookie Says: Left to themselves, things tend to go from bad to worse.