Re: Reverse Compatibility Manifesto

On Fri, 14 Nov 2008 20:19:18 -0500 "William M. Quarles" <walrus(a)bellsouth.net&gt; wrote: The kernel is actually very backward compatible friendly. Very seldom has the kernel itself busted anything (removing support for old style ptys is one instance I can think of). The thing that almost always causes problems is shared libraries being updated to a state of incompatibility, then no "compat" packages being maintained to ship the old libraries. I've often thought it would be better if all software was always shipped with all the shared libs it needs installed in a unique per-program directory, and a system daemon run after each install to determine which libs were actually identical and hard-link them together :-). If virtualization technology keeps on being pushed, perhaps we'll someday reach the point where every program runs inside its very own virtual machine with the exact environment required by that program always maintained in the perfect state for that one program's requirements. At one time it was possible to avoid all these problems by static linking your program, but with every library now being written to dlopen some ridiculous plugin, attempts to static link things usually result in nonsense like dlopen dragging in libc.so to a program which was already linked with libc.a, and pretty soon you have two conflicting copies of malloc stomping on each other (I didn't just make this up - I've watched it happen :-).