svn commit: r216134 - in head: share/man/man9 sys/amd64/include sys/arm/include sys/i386/include sys/ia64/include sys/mips/include sys/pc98/include sys/powerpc/include sys/sparc64/include sys/sun4v...

[John Baldwin]

On Friday, December 03, 2010 9:41:03 am Bruce Cran wrote:
> On Fri, 3 Dec 2010 15:27:13 +0100
> Ivan Voras <ivoras at freebsd.org> wrote:
> 
> > I'd say it depends on if the "specific case that hps raised a few
> > years ago" sentence part refers to an actual problem; i.e. did it
> > happen in practice? If yes, leaving KASSERTs looks like the best
> > option.
> 
> I've found hps's original email thread, which is at
> http://lists.freebsd.org/pipermail/freebsd-hackers/2005-June/012342.html .
> 
> The argument for allowing zero-length reads/writes is at
> http://lists.freebsd.org/pipermail/freebsd-hackers/2005-June/012348.html .
> 
> I suspect the mistake occurred due to a misunderstanding of the API.
> NetBSD have documented it in their man page but not added asserts in
> the code:
> 
> "Functions which take a count of bytes have undefined results if the
> specified count is zero."

I think that is excellent wording and that this case is actually very
rare.

-- 
John Baldwin