Re: [PATCH v3 3/9] dt-bindings: PCI: renesas,pci-rcar-gen2: 'depends-on' is no more optional

[Herve Codina]

Hi Krzysztof,

On Thu, 8 Dec 2022 10:46:32 +0100
Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote:

> On 08/12/2022 10:05, Herve Codina wrote:
> > Hi Krzysztof,
> > 
> > On Thu, 8 Dec 2022 09:26:41 +0100
> > Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote:
> >   
> >> On 07/12/2022 17:24, Herve Codina wrote:  
> >>> The 'depends-on' property is set in involved DTS.
> >>>
> >>> Move it to a required property.
> >>>
> >>> Signed-off-by: Herve Codina <herve.codina@xxxxxxxxxxx>
> >>> ---
> >>>  Documentation/devicetree/bindings/pci/renesas,pci-rcar-gen2.yaml | 1 +    
> >>
> >> This should be squashed with previous patch. There is no point to add
> >> property and immediately in the next patch make it required. Remember
> >> that bindings are separate from DTS.
> >>
> >> Best regards,
> >> Krzysztof
> >>  
> > 
> > I though about make dtbs_check in case of git bisect.  
> 
> And what would this commit change? In Git you will have
> 1. dt-bindings: PCI: renesas,pci-rcar-gen2: Add depends-on for RZ/N1 SoC
> family
> 2. dt-bindings: PCI: renesas,pci-rcar-gen2: 'depends-on' is no more optional
> 
> so what is the difference for git bisect?

Well, today, I have:
1. dt-bindings: Add depends-on
2. dts: Add depends-on
3. dt-bindings: Move depends-on to mandatory

If I squash dt-bindings commits, I am going to have:
  1. dt-bindings: Add mandatory depends-on
  2. dts: Add depends-on
or
  1. dts: Add depends-on
  2. dt-bindings: Add mandatory depends-on

I have not tested but if I used only the first commit in each
case (git bisect):
In the first case, dtbs_check is probably going to signal the
missing 'depends-on' property on dts.
In the second case, dtbs_check is probably going to signal the
not described 'depends-on' property present in dts.

> 
> > 
> > But, ok I will squash or perhaps remove completely this commit.
> > It introduces a DT compatibility break adding a new mandatory
> > property (raised by Rob on cover letter review).
> > Is this compatibility break can be acceptable ?  
> 
> Requiring property in bindings as a fix for something which was broken
> is ok. But this is independent of Linux drivers, which should not stop
> working.

Ok, thanks.

> 
> Best regards,
> Krzysztof
> 

Regards,
Hervé