Philippe Houdoin <philippe.houdoin@xxxxxxx> wrote: > > does anybody the rationale behind going for libnetwork.so? > > Is it only there to be able to have a fresh start? If so, I'd > > prefer > > going the libsocket.so/libbind.so route in our stack - even though > > I > > like libnetwork.so better, I can't stand the code duplication and > > library cluttering for almost nothing. > The best would be to dynamically detect the ABI interface between R5, > BONE and > our ABI at runtime, not compile time. Which would be possible, when we extend the loader's (private) API a bit. > If we want to continue the BeOS Kit way, I don't see why we should > split the > *official* networking kit into several libraries (libnet.so & > libnetapi.so for > R5, libsocket.so, libbind.so & libbnetapi.so for BONE) but just put > them in one > single library: libnet[work].so. Since I'm not that happy with libnetapi.so, I wouldn't mind put that one aside (and eventually put a new one into libnetwork.so some day), but I definitely agree with that idea. > This kind of mess should be 3rd party opportunity only ;-) Even they should leave /system/lib/ alone :-) > > Maybe I should see if our loader could support having both > > libsocket.so > > and libbind.so symlink to the one and only libnetwork.so instead. > That trick works under R5, AFAIK. I use it everytime to switch > between Mesa3D > libGL.so libraries... And I fail to see why we should NOT support it > under > Haiku, whatever the libraries. Network-related or not. It works under BeOS as well as Haiku - but, currently under both systems, the initialization/termination code is called twice (for the same image)! I tested under Linux as well, and there, the loader is smart enough to do that just once :-) Bye, Axel.