It actually does fix a bug because when the value is zero we don't serialize the attributes for the value or mask. This is OK for the value because zero is the default. However, just because the value is zero doesn't mean that the mask can be omitted (and therefore wildcarded).
Longer term, the goal here is to reduce the possibility for these types of bugs by removing exceptions to our normal processing rules. On Sat, Aug 3, 2013 at 2:57 PM, Andy Zhou <az...@nicira.com> wrote: > Jesse, please feel free to correct me if I am wrong. > > As far as I know, this is part of the plan on improving how user space and > kernel are treating missing attributes. > > Missing attribute has to be interpreted to be some default value, and the > value is usually opaque to the other side of the protocol. When a mask is > applied to the field, The opaque value may be matched unintentionally. In > general, we feel that having opaque values will make the netlink protocol > error prone. The saving in netlink bandwidth is not worth the trouble. > > For backward compatibility reasons, some fields, such as in_port and > eth_type, we will continue to allow them to be missing, Priority and > skb_mark, on the hand, can be made explicit and still be backward > compatible, > > Although not fix any bug in particular, this is part the clean up effort. > Kernel side of change is already in. > > > On Sat, Aug 3, 2013 at 2:22 PM, Ben Pfaff <b...@nicira.com> wrote: >> >> The commit message doesn't say why. Why? > > > > _______________________________________________ > dev mailing list > dev@openvswitch.org > http://openvswitch.org/mailman/listinfo/dev > _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
