Re: [RFC PATCH] arm64: dts: qcom: Use plural _gpios node label for PMIC gpios

[Krzysztof Kozlowski]

On 10/12/2022 18:05, Marijn Suijten wrote:
> On 2022-12-10 11:50:51, Krzysztof Kozlowski wrote:
>> On 09/12/2022 23:04, Marijn Suijten wrote:
>>> The gpio node in PMIC dts'es define access to multiple GPIOs.  Most Qcom
>>> PMICs were already using the plural _gpios label to point to this node,
>>> but a few PMICs were left behind including the recently-pulled
>>> pm(i)8950.
>>>
>>> Rename it from *_gpio to *_gpios for pm6125, pm6150(l), pm8005,
>>> pm(i)8950, and pm(i)8998.
>>>
>>> Signed-off-by: Marijn Suijten <marijn.suijten@xxxxxxxxxxxxxx>
>>>
>>> ---
>>>
>>> This was brought up for discussion in [1] but hasn't seen any relevant
>>> reply, unfortunately.
>>
>> This is just a label, it does not matter. Why changing all exisitng
>> files? I don't think it was a part of previous discussions...
> 
> I would've let it slide if it was corrected in the patch that was
> reviewed, but since that didn't happen it wouldn't make sense to only
> correct pmi8950 (and the other bindings submitted or co-authored by
> myself, such as pm6125 and pm8950) for "consistency" - that wouldn't be
> consistent at all.
> 
> To me (and supposedly, to other as well) it does matter.  People keep
> copy-pasting these to to add a newer PMIC and sooner rather than later
> we'll end up with both conventions.
> 
> Regardless, labels are already a mess all over the place, and unless we
> can steer them with bindings or written conventions we're unlikely to
> ever clean that up.
> 
>> To me it is unneeded churn.
> 
> Just like -state and -pins suffix, sometimes even the unnecessary
> -pins-pins suffix?  To me that is the same kind of churn, and *it is

You compare now to node names. The node names have visible impact (name
of devices in sysfs and in system), is affected by DT schema (is parsed)
and should follow DT spec rules.

There is nothing like that about labels, therefore they do not have the
same rules or importance.

I understand that consistency in labels might be good thing... but also
it does not matter. The labels really do not matter, except the code
readability. pmgpio or pmgpios is basically the same readable.

> needed* to keep the bindings somewhat clean, consistent and digestible.
> In this specific case it's not even that many changes, IMO.
> 
> That being said my limited hobby time is probably too valuable to be
> spent on binding cleanup rather than fixes and feature enablement
> elsewhere in the kernel.


Best regards,
Krzysztof