> Now, though, a question:  even though the interface to the libc routines
> hadn't changed, shouldn't such a fundamental interface between libc and
> the kernel have been enough reason to bump the minor version of the
> shared library?

Uh, actually, there were no changes whatsoever to the interfaces to
the libc routines _or_ the kernel.  I.e. in a binary sense, everything
operated exactly as before.

The changes in question simply prototyped __syscall() and dealt with
the problems that that prototype caused (and, for the first rev of the
changes, dealt with those problems badly 8-).


The problem was that __syscall() was being used inconsistently --
sometimes it was being treated as if it returned an off_t, sometimes
it was being treated as if it returned a pointer, sometimes it was
being treated as if it returned an 'int'.

That's actually _correct_.  It does, in fact, return those different
types, depending on the call, rather than returning an off_t.
however, it's impossible to prototype a function to have multiple
return types...

The existing solution is not actually a good solution to the problem,
but it's the best one that can be easily had, in my opinion.



cgd