Re: [PATCH v2 2/6] iio: pressure: Simplify read_* functions

[Vasileios Amoiridis <vassilisamir@gmail.com>]

On Thu, Mar 14, 2024 at 02:32:31PM +0000, Jonathan Cameron wrote:
> On Wed, 13 Mar 2024 21:28:47 +0200
> Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> 
> > On Wed, Mar 13, 2024 at 08:22:45PM +0100, Vasileios Amoiridis wrote:
> > > On Wed, Mar 13, 2024 at 09:01:55PM +0200, Andy Shevchenko wrote:  
> > > > On Wed, Mar 13, 2024 at 06:40:03PM +0100, Vasileios Amoiridis wrote:  
> > 
> > ...
> > 
> > > > >  		case IIO_TEMP:
> > > > > -			ret = data->chip_info->read_temp(data, val, val2);
> > > > > +			ret = data->chip_info->read_temp(data);
> > > > > +			*val = data->chip_info->temp_coeffs[0] * ret;
> > > > > +			*val2 = data->chip_info->temp_coeffs[1];  
> > > >   
> > > > > +			if (!strcmp(indio_dev->name, "bmp580"))
> > > > > +				ret = IIO_VAL_FRACTIONAL_LOG2;
> > > > > +			else
> > > > > +				ret = IIO_VAL_FRACTIONAL;  
> > > > 
> > > > I'm wondering if we may replace these strcmp():s by using enum and respective
> > > > values in chip_info.  
> > > 
> > > The whole problem starts from the fact that all these BMPxxx_CHIP_ID defines are
> > > not unique for the respective BMPxxx device. You mean to add a new variable
> > > that could store some enum values that would be the actual chip_info IDs? Like:
> > > 
> > > enum chip_info_ids = {
> > > 	BMP085,
> > > 	BMP180,
> > > 	...
> > > 	BMP580,
> > > };
> > > 
> > > and later for every chip_info struct to use it like this:
> > > 
> > > const struct bmp280_chip_info bmpxxx_chip_info = {
> > > 	...
> > > 	.chip_info_id = BIT(BMPxxx),  
> > 
> > No BIT(), but yes.
> > 
> Better to use something more meaningful such as just storing the
> type you need to return alongside the values it refers to.
> temp_coeffs_type = IIO_VAL_FRACTIONAL_LOG2 / IIO_VAL_FRACTIONAL as appropriate.
> That way the data and what it is are found in one simple place.
> 
> Basic rule of thumb is that if there is a string comparison to identify
> what to do in a driver (other than deep in the fw handling code) then
> that is a bad design. Likewise any matching on an enum value that correlates
> with that chip ID.  I want to see all the difference between chips in one place,
> not scattered through the code.
> 
> Jonathan
> 

Yes, sounds totally right. I was just hesitating to add new variables in the
chip_info structure (as you probably noticed in the next commits as well).

Best regards,
Vasilis
> 
> > > 	...
> > > }
> > > 
> > > And in the read_raw() function to just use the test_bit() function in the same
> > > way that is done with the test_bit() and avail_scan_mask to test for the
> > > enabled channels?  
> > 
> > If BIT() is more suitable, than also yes.
> > 
>