Re: [RFC PATCH net-next 3/8] tools: ynl: add C-based YNL linter

[Asbjørn Sloth Tønnesen]

On 9/11/26 2:33 AM, Jakub Kicinski wrote:
> On Thu, 10 Sep 2026 22:17:23 +0000 Asbjørn Sloth Tønnesen wrote:
> > Validate that structures described in YNL match their C counterpart,
> > currently focused on enums.
>
> I don't see the point of this. LLMs will catch most of the problems
> and for the hundredth time classic netlink is not the focus for YNL.
> Anything genetlink can have YNL generate the code and uAPI so problems
> do no exist.

FTR, I didn't intent to imply any "regular" C code gen on classic families,
the --linter was a very limited output mode, only for validating enum alignment.

I like to also have deterministic checks, the LLM didn't catch the
IPV4_DEVCONF_* alignment issue, while reviewing commit 5e63ade19044
("netlink: specs: rt-link: update ipv6 devconf doc"), where I claimed
that IPV4_DEVCONF_FORWARDING = 1 (while it was 0 in YNL) in patch 1/3.

Now reading the raw log, the closest it got was "... which means there could
be an existing bug in `ipv4-devconf`!", but then it forgot about it, and
that was for patch 3/3. I can't find it in the Clashiko log for that series.

https://sashiko.dev/#/log/148492

> Feel free to keep the code OOT and let's see if you can prove my
> assertion wrong within a couple of release :)

I will keep it OOT for now. Thank you for your review!

> If you have any ready fixes to the spec please post them.
> I'm also sitting on a bunch of fixes based on LLM scans, I'm just
> blocked on the "15 outstanding patches" limit :(

I have posted my fixes for most severe issues now, and left out adding
extra enum-prefix and header entries. I skimmed your new devlink spec
patches, to look for duplicates, but will do a review tomorrow.