Re: [PATCH] dt-bindings: iio: bmc150_magn: Document missing compatibles

[Jonathan Cameron <jic23@kernel.org>]

On Mon, 29 Jun 2020 08:49:25 +0200
Krzysztof Kozlowski <krzk@kernel.org> wrote:

> On Sat, Jun 27, 2020 at 03:57:14PM +0100, Jonathan Cameron wrote:
> > On Mon, 22 Jun 2020 07:19:40 +0200
> > Krzysztof Kozlowski <krzk@kernel.org> wrote:
> >   
> > > On Sat, Jun 20, 2020 at 04:40:49PM +0100, Jonathan Cameron wrote:  
> > > > On Wed, 17 Jun 2020 12:12:59 +0200
> > > > Krzysztof Kozlowski <krzk@kernel.org> wrote:
> > > >     
> > > > > The driver supports also BMC156B and BMM150B so document the compatibles
> > > > > for these devices.
> > > > > 
> > > > > Fixes: 9d75db36df14 ("iio: magn: Add support for BMM150 magnetometer")
> > > > > Cc: <stable@vger.kernel.org>
> > > > > Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
> > > > > 
> > > > > ---
> > > > > 
> > > > > The fixes tag is not accurate but at least offer some backporting.    
> > > > 
> > > > I'm not sure we generally bother backporting a missing section of binding
> > > > documentation. Particularly as this doc isn't in yaml yet so it's not
> > > > as though any automated checking is likely to be occurring.
> > > > 
> > > > Rob, any views on backporting this sort of missing id addition?
> > > > 
> > > > One side comment here is that the devices that are magnetometers only
> > > > should never have had the _magn prefix in their compatibles. We only
> > > > do that for devices in incorporating several sensors in one package
> > > > (like the bmc150) where we have multiple drivers for the different
> > > > sensors incorporated. We are too late to fix that now though.  It
> > > > may make sense to mark the _magn variants deprecated though and
> > > > add the ones without the _magn postfix.    
> > > 
> > > I can add proper compatibles and mark these as deprecated but actually
> > > the driver should not have additional compatibles in first place - all
> > > devices are just compatible with bosch,bmc150.  
> > 
> > Why not?  Whilst the devices may be compatible in theory, it's not unusual
> > for subtle differences to emerge later.   As such we tend to at least
> > support the most specific compatible possible for a part - though we
> > can use fallback compatibles.  
> 
> It does not strictly harm but have in mind that adding is always
> possible (when you spot the difference between devices). But it is
> entirely different with removal - it takes time to deprecate one and to
> remove it.
> 
> There is just no benefit for adding new compatibles for really
> compatible devices. The module device table just grows. It makes sense
> however to document in bindings that given compatible serves family of
> devices.
> 
> Somehow driver developers got impression that they need to make a commit
> like "Add support for xyz123 device" adding only compatible, to bring
> support for new device. But the support was already there so just
> document that xyz001 is compatible with xyz123.

Whilst I agree the compatible is not really necessary, it is often
non trivial to establish two parts are actual compatible. Manufacturers
have an annoying habit of not actually saying so on their datasheets.
So it is useful to add documentation for the support so that a grep
will identify the driver supports it.

I don't have a problem with people adding the ID particularly as they
are often not entirely sure the parts are compatible.  I'm not fussed
if they don't do so of course.

Ideal in my view is to list multiple compatibles in the dts files
in this case to allow us to support any differences if any turn up
in the future. 

From a purely practical basis, if I'm writing a DTS I'd much rather
it matched up with my BOM rather than having to 'know' that two parts
are compatible.

Jonathan

> 
> Best regards,
> Krzysztof
>