Re: CVS commit: src/sbin/ifconfig

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Wed, Sep 14, 2016 at 02:17:26PM +0100, Roy Marples wrote:
> On 14/09/2016 13:00, Joerg Sonnenberger wrote:
> > On Wed, Sep 14, 2016 at 11:46:43AM +0000, Roy Marples wrote:
> >> Module Name:	src
> >> Committed By:	roy
> >> Date:		Wed Sep 14 11:46:43 UTC 2016
> >>
> >> Modified Files:
> >> 	src/sbin/ifconfig: media.c
> >>
> >> Log Message:
> >> Don't bail if SIOGIFMEDIA doesn't return any media lists because we
> >> can still report link status.
> > 
> > As I mentioned in the review, I think this is the wrong approach. I find
> > the assumption that every interface supporting the IFMEDIA interface has
> > at least one medium to be perfectly reasonable. Setting up and
> > defaulting to one AUTO medium is two lines of code. Now we have to look
> > for bugs in every consumer of this interface...
> 
> There is no media to select because there is none. Providing a list of
> one just to satisfy strikes me as waste of bytes.

Of course there is a medium. You are transporting the byte stream via
something. Now technically, this can be Ethernet or a serial line or
whatever and that's not really relevant, so using something like
IFM_GENERIC would certainly be fine.

> Existing code that just cares about link status will function as before
> without any change.
> Ditto for any code which lists or sets the media in use because there
> has to be a list to start with.

You are only looking at code in the tree. Things like net-snmp and other
monitoring/reporting tools have a long list of strange assumptions. It
doesn't strike me as useful to make the job for them more difficult
without reason.

> It could also be argued that obtaining link status should be it's own
> ioctl and not bolted on-to media.

Separate issue, but it certainly sounds useful enough.

Joerg