soren@wheel.dk ("Soren S. Jorvang") writes:
> it seems there is a p{read,write}v implementation in HP-UX that
> differs somewhat from ours.

are there any standards which specify a p{read,write}v?  a web search
makes it look like the answer is 'no'.


> In particular, it has the per-iovec offsets.

meaning, each iovec has an offset parameter?  ("the" confuses me.)


> I think we should probably alter ours to match. What to do
> about compatibility (our version was in 1.4)? Is it even used
> by anything/anyone?

certainly, there's something to be said for either behaviour; the
behaviour you want is determined by your application...  I don't know
if anything uses the one we currently provide.  I can think of some
reasons you might want to use it, certainly.


at minimum, it sounds like somebody should inform the iozone people
that not all p{read,write}v implementations -- of which there are two,
it seems -- behave the same.

if we provided it as a supported interface (and we did), then you need
to provide backward compatibility just like you would for any other
standard interface that's changed.  Also, if there's no compelling
reason to remove the interface, i'd suggest keeping it as-is and
adding a new interface to do the right thing, if desired.  (and,
fixing iozone to do the right thing.)



cgd