Re: [PATCH 1/5] power: RFC: introduce a new power API

[Andres Salomon <dilinger@queued.net>]

On Wed, 19 Dec 2007 15:35:46 +0300
Anton Vorontsov <cbou@mail.ru> wrote:

> On Tue, Dec 18, 2007 at 02:10:01AM -0500, Andres Salomon wrote:
> > On Mon, 17 Dec 2007 14:24:16 +0300
> > Anton Vorontsov <cbou@mail.ru> wrote:
> > 
> > > On Mon, Dec 17, 2007 at 02:41:39AM -0500, Andres Salomon wrote:
> > > [...]
> > > > > > On Sun, 2007-12-16 at 21:24 -0500, Andres Salomon wrote:
> > > > > > > This API has the power_supply drivers device their own device_attribute
> > > > > > > list; I find this to be a lot more flexible and cleaner.  
> > > > > 
> > > > > I don't see how this is more flexible and cleaner. See below.
> > > > > 
> > > > > > > For example,
> > > > > > > rather than having a function with a huge switch statement (as olpc_battery
> > > > > > > currently has), we have separate callback functions.
> > > > > 
> > > > > Is this an improvement? Look into ds2760_battery.c. I scared to
> > > > > imagine what it will look like after conversion.
> > > > 
> > [...]
> > > 
> > > I see your point now. Basically, now I'm encourage to think just one
> > > more time: is there third (better) option in addition to current and
> > > this? I still hope there is some not obvious, but elegant solution.
> > > If there isn't, I'm ready to surrender and will help with everything
> > > I can.
> > > 
> > 
> > Hm.  It occurs to me that there's nothing keeping us from having a
> > single callback for the driver properties.  Keeping the other patches
> > the same, do you prefer the following approach versus what was originally
> > in patch#3?
> 
> Why so difficult? Maybe like this:
> 

The point is to get rid of 'propval', and having the core driver define
formats.  That's one of the places where we ran into problems with the
current API; by having the core driver define what type a property should
be returning, we limit battery drivers to what they can display, as well
as encourage a lot of non-shared code to end up in the core driver.  That's
the reason why we strcpy into 'buf', rather than val->strval.

For transitioning, we could certainly just use val->strval all of the time,
but there's not much point in doing that in the long term; we might as well
just pass around 'buf'.


> diff --git a/drivers/power/olpc_battery.c b/drivers/power/olpc_battery.c
> index c998e68..00f0b71 100644
> --- a/drivers/power/olpc_battery.c
> +++ b/drivers/power/olpc_battery.c
> @@ -176,13 +176,13 @@ static int olpc_bat_get_property(struct power_supply *psy,
>  
>  		switch (ec_byte >> 4) {
>  		case 1:
> -			val->strval = "Gold Peak";
> +			ret = sprintf(val->strval, "%s\n", "Gold Peak");
>  			break;
>  		case 2:
> -			val->strval = "BYD";
> +			ret = sprintf(val->strval, "%s\n", "BYD");
>  			break;
>  		default:
> -			val->strval = "Unknown";
> +			ret = sprintf(val->strval, "%s\n", "Unknown");
>  			break;
>  		}
>  		break;
> diff --git a/drivers/power/power_supply_sysfs.c b/drivers/power/power_supply_sysfs.c
> index 249f61b..83e127d 100644
> --- a/drivers/power/power_supply_sysfs.c
> +++ b/drivers/power/power_supply_sysfs.c
> @@ -54,7 +54,9 @@ static ssize_t power_supply_show_property(struct device *dev,
>  	ssize_t ret;
>  	struct power_supply *psy = dev_get_drvdata(dev);
>  	const ptrdiff_t off = attr - power_supply_attrs;
> -	union power_supply_propval value;
> +	union power_supply_propval value = {
> +		.strval = buf,
> +	};
>  
>  	ret = psy->get_property(psy, off, &value);
>  
> @@ -75,7 +77,7 @@ static ssize_t power_supply_show_property(struct device *dev,
>  		return sprintf(buf, "%s\n",
>  			       capacity_level_text[value.intval]);
>  	else if (off >= POWER_SUPPLY_PROP_MODEL_NAME)
> -		return sprintf(buf, "%s\n", value.strval);
> +		return ret;
>  
>  	return sprintf(buf, "%d\n", value.intval);
>  }