tech-toolchain archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: ar "zero" flag



On Thu, Aug 14, 2008 at 10:28:45PM -0700, Jason Thorpe wrote:
>> - Desupporting static binaries. This would move our compat abi promise
>>   from kern<->userland to (basically) libc<->appliation. Think windows.
>>   This might make some issues easier, but to be effective, we would
>>   have to use the big time machine, change history and do this effectively
>>   before NetBSD 0.9 happend, or drop our existing compat promises - IMHO
>>   this proposal is unlikely to reach quorum.
>
> Windows is not the only platform that does this.  OS X is like this, as is 
> Solaris, IIRC.  We don't need a big time machine to go back and fix 0.9 ... 
> We simply need to make the decision, and from that point forward, no more 
> need for new compat code in the kernel.  When we bump the libc major is a 
> good time for this, I think.

Since nobody has said this yet so far: moving the compatible ABI from
the kernel to libc would be an extremely bad move all on its own,
regardless of static vs. dynamic linking.

The kernel ABI is reasonably small, well documented, and well
understood. The mechanisms for providing compatibility are also well
understood and well tested. Furthermore, by its inherent nature the
kernel ABI is in large part explicitly specified and manually
maintained. These properties make it robust. The libc ABI is none of
these things. It is not robust. It's a lot easier to break by
accident.

(And currently, we don't even have adequate mechanisms for providing
compatibility in shared libraries.)

-- 
David A. Holland
dholland%netbsd.org@localhost


Home | Main Index | Thread Index | Old Index