Re: [RFC net-next 0/4] ynl: YAML netlink protocol descriptions

[sdf@google.com]

On 08/11, Jakub Kicinski wrote:
> Randomly adding Michal to CC since I just realized I forgot
> to CC him on the series.

> On Thu, 11 Aug 2022 08:01:52 -0700 Stephen Hemminger wrote:
> > > On Wed, 10 Aug 2022 21:15:34 -0700 Stephen Hemminger wrote:
> > > > Would rather this be part of iproute2 rather than requiring it
> > > > to be maintained separately and part of the kernel tree.
> > >
> > > I don't understand what you're trying to say. What is "this",
> > > what is "separate" from what?
> >
> > I am saying that ynl could live as a standalone project or as
> > part of the iproute2 tools collection.

> It's a bit of a strange beast, because the YNL C library ends up being
> relatively small:

>   tools/net/ynl/lib/ynl.c                  | 528 +++++++++++++++++++++++++
>   tools/net/ynl/lib/ynl.h                  | 112 ++++++

> The logic is mostly in the codegen:

>   gen.py                                   | 1601 +++++++++++++++++++++++++

> but that part we need for kernel C code as well.

> The generated code is largish:

>   tools/net/ynl/generated/dpll-user.c      | 371 ++++++++++++++++++
>   tools/net/ynl/generated/dpll-user.h      | 204 ++++++++++
>   tools/net/ynl/generated/ethtool-user.c   | 367 ++++++++++++++++++
>   tools/net/ynl/generated/ethtool-user.h   | 190 +++++++++
>   tools/net/ynl/generated/fou-user.c       | 322 ++++++++++++++++
>   tools/net/ynl/generated/fou-user.h       | 287 ++++++++++++++
>   tools/net/ynl/generated/genetlink-user.c | 635  
> +++++++++++++++++++++++++++++++
>   tools/net/ynl/generated/genetlink-user.h | 201 ++++++++++

> but we don't have to commit it, it can be created on the fly
> (for instance when a selftest wants to make use of YNL).

> Then again it would feel a lot cleaner for the user space library
> to be a separate project. I've been putting off thinking about the
> distribution until I'm done coding, TBH. Dunno.

my 2c:

Putting it into iproute2 will make it carry a 'networking' badge on it
meaning no other subsystem would look into it.

I'd rather make netlink more generic (s/netlink/kernelink/?) and remove
CONFIG_NET dependency to address https://lwn.net/Articles/897202/