I think I'd like it in cfm.h better, if that's OK.
On Wed, Mar 23, 2011 at 11:19:40AM -0700, Ethan Jackson wrote: > The main issue is that I wanted to pull L2 header composition out of > CFM into it's caller. This will allow us to remove the eth_src from > the cfm configuration and prevent the ugliness which occurs when we > push the eth_src from the bridge to ofproto through the cfm > configuration. We could alternatively put it in the cfm header file > if you prefer. > > Ethan > > On Wed, Mar 23, 2011 at 10:00 AM, Ben Pfaff <[email protected]> wrote: > > On Tue, Mar 22, 2011 at 06:32:53PM -0700, Ethan Jackson wrote: > >> This seems to be more stylistically correct. > > > > Is there any more rationale than this? ?It's never been my goal to > > have *every* protocol definition in packets.h, although I can see how > > you got that idea. ?We used to have better counterexamples than we do > > now, in the lib/dhcp.h and lib/stp.h libraries. > > > > I can think of a few reasons to put declarations in packets.h instead > > of in a protocol-specific header: > > > > ? ? ? ?- The declarations are widely used and so there's little > > ? ? ? ? ?benefit in making the users include additional headers. > > > > ? ? ? ?- There's a clean logical separation in our tree between a > > ? ? ? ? ?protocol and implementations of it, but it doesn't seem > > ? ? ? ? ?worthwhile (yet) to make a new header file for just this > > ? ? ? ? ?protocol. > > > > ? ? ? ?- There just isn't another good place in the tree to put the > > ? ? ? ? ?declarations. > > > > But I don't know which of these applies to cfm. ?It seems that, so far > > at least, only cfm.c really needs these definitions, and so cfm.c > > seems like an OK place, to me. > > _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
