> I think it's at least confusing that the native emulation (NetBSD)
> needs to set a flag to specify its previous behaviour.  The logic of
> this seems backwards to me; I'd rather expect emulations to specify
> that they do _not_ want the native behaviour.

And Jason R Thorpe said:

> This is wrong!  I said *specifically* that the sense of the flag should be
> the other way, i.e. a flag to *disable* the BSD pipe async i/o semantics.

Well NetBSD behavior is nearly an exception. The only OSes behaving the
same way are Digital UNIX and Darwin. Even FreeBSD and OpenBSD do not do
it the way we do. And this is where the problem is: we are about to
import FreeBSD pipes that do not implement our current semantics. 

I used the flag in that way, because our native semantics will soon
correspond to no flag. Considering the new pipe problem, do we still
want it to be inverted?

-- 
Emmanuel Dreyfus
p99dreyf@criens.u-psud.fr