Re: [PATCH v3 5/7] dt-bindings: iio: bosch,bme680: Add supply properties

[Rob Herring <robh@kernel.org>]

On Sun, Nov 03, 2024 at 10:46:46AM +0100, Krzysztof Kozlowski wrote:
> On Sat, Nov 02, 2024 at 03:33:15PM +0000, Jonathan Cameron wrote:
> > On Sat,  2 Nov 2024 14:13:09 +0100
> > Vasileios Amoiridis <vassilisamir@gmail.com> wrote:
> > 
> > > Extend dt-binding for BME680 gas sensor device. The device incorporates
> > > as well temperature, pressure and relative humidity sensors.
> > This description should make it clear it is moving from trivial-devices.yaml
> > 
> > dt-bindings: iio: bosch,bme680: Move from trivial-bindings and add missing supplies.
> > 
> > Then say a little more on why you are moving it.
> > 
> > > 
> > > Signed-off-by: Vasileios Amoiridis <vassilisamir@gmail.com>
> > 
> > There was an open question on the previous version about
> > setting the supplies as required (which I see you've removed).
> > My understanding previously was that it is fine to make that change
> > in a binding if it reflects supplies that are required to be enabled
> > for the device to function at all.  If there were previously missing
> > that's a binding bug we should fix.
> > 
> > I'd like a clarification from the DT binding maintainers on that.
> > Obviously doesn't work for other users of dt bindings but in
> > Linux this would be fine as they were already on for any board
> > that worked and the regulator framework will through us a fake
> > regulator for cases like this.
> > 
> > https://lore.kernel.org/all/20241022182451.00007ac0@Huawei.com/
> > 
> > Jonathan
> 
> That was Rob's objection so I will leave it to him, but putting my two
> cents in for Linux it is not an ABI break because missing regulator
> supplies are substituted with dummy ones. Unless something changed...

Shrug. I don't think we're entirely consistent on this. If we're saying 
supplies are always required, then every device in trivial-devices.yaml 
is wrong. Since Linux handles them missing, you can also argue that 
supplies are never required.

I'd prefer not to special case regulators as an exception I have to 
remember. I have some rudimentary ABI checking I'm working on that 
checks for things like new required properties. Though it wouldn't catch 
this particular change given it moves the schema.

Rob