> >>whoah Marcus, calm down, stop seeing conspiracys everywhere. >>im not trying to start a flame war, and im sure neither is helmar. > >We really are all on the same side here. Please, everyone, treat each >other kindly. > >>there have been a few compelling reasons for breaking binary >>compatibility mentioned back when R5 was still being worked on be >>be,inc (gcc 3.0, better kernel virtual memory , performance issues, >>etc), so it might not be a bad idea to consider doing so if there are > >Of these, only GCC requires a real binary compatability break. And that >is only maybe. > >>compelling reasons for us to do so. yes, it would be a good thing to >>keep compatability if possible, but it should not be so sacred that we >>get into the mess of trying to keep compat with a 20 year old os. ;) >>Be,inc completly broke binary compat once, and there were minor breaks >>each release. > >The only decision we have reached on this line is that R1 will *NOT* break >compatability unless it absolutely has to. > >>that being said, if you can keep binary compat please do, it would make >>everyones life simpler. i will be trying to. its really the kernel/IK >>people that will make or break it. > >The facts, as I see them, are this: > >binary compatability means that the kernel can load an R5 app. Link it to appropriate .so's. And execute it. >It has *nothing* to do with the internal workings of the kits. Only the interface (i.e. things like number of virtual functions) >must remain the same. That is why GCC3 breaks bin compatability - they reworked the whole way that the >binaries look. That is *not* to say that it would be impossible to have OBOS read *both*. Caveat - I have not looked >in depth at GCC 3. The ELF loader is generic. The linking mechanism should also work - the names are different, but the >methodology should work. In fact, the way I see it, we could even build libbe.so (compatability version) and libbe.so (new version with GCC3). New apps, that know about the new version would load it. Note that I didn't indicate the mechanism for determining. That is alwasy flamewar fodder. :-) > How about libbe.so.1 etc., like in the UN*X world? Would definitely make shared libraries easier to manage (both system-wide and user), for situations like these. -- Mikael J @ http://hem.spray.se/tic_khr