[ovs-dev] [classifier-opt 19/28] flow: Replace flow_wildcards members by a single "struct flow".

[Ben Pfaff]

One more thought: it's pretty easy to get rid of flow_wildcards later
if we don't want it; it's harder to reintroduce it.

On Tue, Aug 07, 2012 at 01:45:04PM -0700, Ethan Jackson wrote:
> Sounds reasonable.
> 
> Ethan
> 
> On Tue, Aug 7, 2012 at 1:41 PM, Ben Pfaff <blp at nicira.com> wrote:
> > On Mon, Jul 30, 2012 at 06:10:22PM -0700, Ethan Jackson wrote:
> >> In flow_wildcards_init_catchall() and flow_wildcards_init_exact() why
> >> not just memset?  Perhaps the appropriate time to make that change
> >> would have been when in_port was changed to a mask come to think of
> >> it.  It's fine to leave it if you're going to resolve it in a future
> >> patch of the series.
> >
> > That (and other simplifications) happen in patch 21/28, so I guess
> > I'll leave them there.
> >
> >> The indentation isn't quite right in flow_wildcards_combine().  It was
> >> incorrect before this patch as well, but this may be a good time to
> >> clean it up.
> >
> > OK, fixed.
> >
> >> I suspect you're going to switch flow_wildcards_equal() to using
> >> memcmp() in a future patch?
> >
> > Yes.
> >
> >> Do we still need "struct flow_wildcards" at all?  We could just use
> >> struct flow directly.  Again, perhaps this will make more sense once
> >> I've seen the future patches.
> >
> > We don't need flow_wildcards.  I kept it on the notion that it was
> > useful to readers of code to be able to distinguish a flow from a set
> > of wildcards for a flow, and useful from a type system perspective for
> > the same reason.  I may be wrong; I don't know.