On Thu, 2008-10-30 at 14:47 -0400, Peter Memishian wrote:
> > >   * If we're going to include the ip_*register_hook() API, why do
>  > >    we omit ipobs_init(), ipobs_fini() and ipobs_hook()?
>  > 
>  > Because those aren't an API, but functions internal to the ip module.
> 
> They're still an API even if they're internal.  Personally, I don't see
> much difference between ipnet.c calling ipobs_register_hook() and tcp.c
> calling ipobs_hook(); having tcp.c compiled into ip but ipnet.c depend on
> ip seems an implementation matter, not an architectural one.

That's true.  There's also an architectural distinction between the two
sets of functions.  One set of function allows a consumer access to IP
packets.  The other set of functions helps the provider implement access
to these IP packets for the consumers.

Let's just include them, I don't really care which project-private
functions appear in these materials, to be honest.

>  > >  * Why are we classifying the net_* and NE_* stuff as Committed?
>  > 
>  > That is because the rest of the netinfo API is Committed.
> 
> I find that a weak justification.  Just because part of an API is
> Committed doesn't mean all parts of it need to be.

That's true, but I also don't have a good justification for providing
access to interface flags and zoneid's, but making it private while
every other aspect of the API is Committed.  Is there a particular
concern?

-Seb



Reply via email to