Re: [PATCH v1 2/6] device property: Add fwnode_property_match_property_string()

[Jonathan Cameron <jic23@kernel.org>]

On Thu, 10 Aug 2023 16:26:54 +0300
Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:

> On Wed, Aug 09, 2023 at 06:59:44PM +0100, Jonathan Cameron wrote:
> > On Tue,  8 Aug 2023 19:27:56 +0300
> > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:  
> 
> ...
> 
> > > +int fwnode_property_match_property_string(const struct fwnode_handle *fwnode,
> > > +	const char *propname, const char * const *array, size_t n)  
> > 
> > Hi Andy,
> > 
> > Whilst I'm not 100% sold on adding ever increasing complexity to what we
> > match, this one feels like a common enough thing to be worth providing.  
> 
> Yep, that's why I considered it's good to add (and because of new comers).
> 
> > Looking at the usecases I wonder if it would be better to pass in
> > an unsigned int *ret which is only updated on a match?  
> 
> So the question is here are we going to match (pun intended) the prototype to
> the device_property_match*() family of functions or to device_property_read_*()
> one. If the latter, this has to be renamed, but then it probably will contradict
> the semantics as we are _matching_ against something and not just _reading_
> something.
> 
> That said, do you agree that current implementation is (slightly) better from
> these aspects? Anyway, look at the below.
> 
> > That way the common properties approach of not checking the return value
> > if we have an optional property would apply.
> > 
> > e.g. patch 3  
> 
> Only?
I didn't look further :)

> 
> > would end up with a block that looks like:
> > 
> > 	st->input_mode = ADMV1014_IQ_MODE;
> > 	device_property_match_property_string(&spi->dev, "adi,input-mode",
> > 					      input_mode_names,
> > 					      ARRAY_SIZE(input_mode_names),
> > 					      &st->input_mode);
> > 
> > Only neat and tidy if the thing being optionally read into is an unsigned int
> > though (otherwise you still need a local variable)  
> 
> We also can have a hybrid variant, returning in both sides
> 
>   int device_property_match_property_string(..., size_t *index)
>   {
> 	  if (index)
> 		  *index = ret;
> 	  return ret;
>   }
> 
> (also note the correct return type as it has to match to @n).
> 
> Would it be still okay or too over engineered?
> 
Probably over engineered....

Lets stick to what you have.  If various firmware folk are happy with
the new function that's fine by me.  Rafael?

Jonathan