On Tue, Nov 25, 2014 at 05:08:23PM -0800, Ben Pfaff wrote: > On Thu, Nov 20, 2014 at 04:42:53PM -0800, Jarno Rajahalme wrote: > > All the users of miniflow extract supply a miniflow with inlined data. > > Make the extraction a bit more efficient by only supporting this. > > > > Also, clean up miniflow building utilities in preparation for moving > > them to lib/flow.h in a later patch. > > > > Signed-off-by: Jarno Rajahalme <jrajaha...@nicira.com> > > Usually I just write a "module" name in the subject, e.g. "flow: Only > support inline values..." I'm not sure that it's necessary to mention > that it's a .c or .h file. (I see that a few patch subjects in this > series are written that way.) > > This code for direct construction stuff is complex. I wonder whether > a simpler approach would be just as fast and easier to understand. > Suppose that, instead of constructing a miniflow initially, we instead > construct a regular "struct flow" on the stack. All the zeroing and > then later checking for nonzero values is what drove us earlier to > move to building a miniflow directly, so we'd want to avoid that. But > we can do that by not initializing the flow at all, and just keeping > track in a map of the u32s we've initialized, and then copying those > fields into a miniflow based on the map we assembled.
I'm not sure of the status of this series. Some of the patches have been applied, but I don't think there was any further discussion of my comment here, and I don't think that there's a v2 of the series. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
