Re: [PATCH v1 2/3] net/hyperv: implement core functionality

[Stephen Hemminger <stephen@networkplumber.org>]

On Mon, 18 Dec 2017 22:03:55 +0100
Thomas Monjalon <thomas@monjalon.net> wrote:

> > 
> > Good question. For the following reasons:
> > 
> > - I forgot about the existence of ether_ntoa() and didn't look it up seeing
> >   struct ether_addr is (re-)defined by rte_ether.h. What happens when one
> >   includes netinet/ether.h together with that file results in various
> >   conflicts that trigger a compilation error. This problem should be
> >   addressed first.
> > 
> > - ether_ntoa() returns a static buffer and is not reentrant, ether_ntoa_r()
> >   is but as a GNU extension, I'm not sure it exists on other OSes. Even if
> >   this driver is currently targeted at Linux, this is likely not the case
> >   for other DPDK code relying on rte_ether.h.
> > 
> > - I had ether_addr_from_str()'s code already ready and lying around for a
> >   future update in testpmd's flow command parser. No other MAC-48 conversion
> >   function I know of is as flexible as this version. The ability to omit ":"
> >   and entering partial addresses is a big plus IMO.
> > 
> > I think both can coexist on their own merits. Since rte_ether.h needs to be
> > fixed either way, how about I move this function in a separate commit and
> > address the conflict with netinet/ether.h while there?  
> 
> Looks to be a good plan.

Agree, rte_ether is where it should go. Please put functions for parsing there.
The name and logic conflict between netinet/ether.h and rte is both a blessing
and a curse. Although the definitions of ether_addr overlap, they are equivalent.