On Wed, 24 Jan 2018 09:52:16 +0100, Jiri Pirko wrote:
> >@@ -606,9 +606,9 @@ struct tc_cls_common_offload {
> > };
> >
> > static inline void
> >-tc_cls_common_offload_init(struct tc_cls_common_offload *cls_common,
> >- const struct tcf_proto *tp,
> >- struct netlink_ext_ack *extack)
> >+tc_cls_common_offload_init__(struct tc_cls_common_offload *cls_common,
> >+ const struct tcf_proto *tp,
> >+ struct netlink_ext_ack *extack)
>
> This is really confusing. You do duplication and this odd function
> naming only to remove it at the end of the patchset.
> But the only goal is to add flags arg and tc_skip_sw check.
>
> So just add the arg and check and change all the calls to this finction
> to pass the flags arg. You will reduce like 8 patches into one that
> is well-readable.
I'm trying to keep functional and non-functional changes separate.
I talked to Simon and he also prefers to keep things separate. Now
classifier maintainers can look at their classifier easily without
drowning in a soup of tens of almost-identical patch hunks.
Look at the u32 classifier where I have to save the flags. It would
be hard to notice that in 20-hunk patch.
Does anyone else have a preference? It's easy enough to squash the
patches...
