Re: [PATCH net-next] Documentation: networking: Add a test plan for ethtool pause validation

[Xuan Zhuo <xuanzhuo@linux.alibaba.com>]

On Wed, 27 May 2026 14:08:30 +0200, Andrew Lunn <andrew@lunn.ch> wrote:
> > As for the kAPI testing, I agree that the end goal is to get driver
> > authors to get their flow control implementation right running this
> > suite.
> >
> > But I don't really see how we can validate kAPI itself, as we're down at
> > the ethnl level.
>
> All we can do is invoke the kAPI in different ways, and test we get
> the expected results. When it fails, it is down to the developer to
> figure out why, which layer. But they have a description of what the
> test is doing, and why? In most reviews, all i need to explain is the
> expected behaviour, and the second version is correct. So a test with
> explanation text should sort cut that process. I don't think we need
> any more.
>
> To some extent, we have an iterative process here. We have never done
> testing of this, we don't know exactly what we need. If we get
> feedback that a test is failing, but they cannot figure out why, we
> might need to help out, and then extend either the text, or add finer
> grain testing to narrow down the problem space. If we get a submission
> which passes all the tests but review turns up problems, we might want
> to ask the developers to extend the tests to catch the failure.
>

So I've been thinking lately: should we let AI generate and maintain these tests,
including kselftest? This would give us a much richer and more comprehensive
set of tests. Plus, each test could come with a complete explanation of its
purpose and methodology. In short, much of the work we used to do manually
can be offloaded to AI. This way, we can build a massive test suite and
achieve much broader coverage.

I'm actually trying this out right now.

Thanks.


>     Andrew
>