On Wed, Sep 29, 2021 at 7:08 PM [off-list ref] wrote:

[sending to you directly because I am no longer on kernel list and you
respond to the dtschema queries anyway, if you prefer I will re-join
kernel list and re-send to whole list]

Reviving this thread now that we have some experience with dt-schema.

Our experience with both dtc and the dt-schema scripts is there doesn't
seem to be any real distinction between errors and warnings.

Everything is an error, but there are too many (for dtbs) or
introduced too frequently (for bindings), so they are all warnings.

What I really want to distinguish is warnings/errors for new
bindings/users vs. existing ones. Things we already have and can't
change, but don't want new ones.

Below are some examples.

This is from dt-validate: : pinctrl@f100000: 'width' is a required
property
This is from dtc: : Warning (reg_format): /soc/pinctrl@f100000:reg:
property has invalid length (8 bytes) (#address-cells == 2, #size-cells
== 1)

In both cases neither tool returned an error code to the shell (echo $?
= 0)
dtc will error with a syntax problem, but that seems to be it.

Is this how the kernel community prefers these tools to work?

The standard for the kernel is no warnings in general (perhaps you saw
the recent -Werror change). So it's really only a question of erroring
out or not.

Both submitters and maintainers frequently don't run the schema checks
and things get applied. It's better than it was, but still happens. If
we errored out, those that do run checks would always have a broken
tree. In fact, we did error out in some cases and I made it not. I
always run 'make -k', but it seems many don't and miss testing their
stuff due to erroring out.

dtc could probably be changed to error out at least for 'dtbs' target.
Though it would need a -Werror type option or you'd have to list every
warning option on the command line.

Our concern is more with the dtschema scripts so we can use this to
break the build and force the engineer to fix either the .yaml or .dtsi
file.

Before we dive into the dtschema scripts we wanted to understand the
philosophy behind the design decisions.


On 2020-11-30 11:02, Rob Herring wrote:

quoted

On Wed, Nov 25, 2020 at 2:13 PM [off-list ref] wrote:

quoted

Folks,
I am advocating use of dt-schema project internally to validate DTS
files.
I should add that our use is outside kernel tree on proprietary
project.

One of the push-backs I'm getting from the management chain is along
the
lines of...

Who is this Rob Herring guy and why should we use a project that is
only
sourced on https://github.com/robherring/dt-schema?

I wouldn't trust him...

That's the wrong repo though:
https://github.com/devicetree-org/dt-schema

(Unfortunately, GH's forks is misleading as the 'root' repo has
changed.)

quoted

If the kernel project is using it, why isn't kernel.org hosting the
project?

It's not a kernel project. That's why devicetree.org hosts it.

quoted

What is kernel plan if Rob walks away from the project, is this going
to
wither away and die?

IMO, only if folks don't find validation valuable or a better
implementation comes along.

quoted

There are more, but the above pseudo-quotes grab the gist of the
management complaints.

Q.1) Is there a plan for the kernel project to suck dt-schema into its
orbit?

No, the 'plan' (more like desire) is more in the opposite direction.
Move more of DT (bindings and dts files) out of the kernel for other
projects to use. For now, we have the 'devicetree-rebasing' tree which
is just the DT bits from the kernel tree.

quoted

Q.2) How many active maintainers are there for dt-schema?

Mostly just me. Maxime Ripard is also one. Others could be if the need
arose.

quoted

Q.3) How do I respond to the above types of complaints?

jsonschema python module which is our main dependency is also just a
single maintainer. So is dtc. Maybe not what you want to highlight.

Rob