Re: [PATCH 1/2] platform: make platform_get_irq_optional() optional (summary)

[Uwe Kleine-König]

Hello,

I'm trying to objectively summarize the discussions in this thread in
the hope this helps finding a way that most people can live with.

First a description of the status quo:

There are several function pairs *get() and *get_optional() that however
are different in various aspects. Their relevant properties are listes
in the following table. Ideally each line had only identical entries.

					| clk_get		| gpiod_get		| platform_get_irq	| regulator_get		|
	return value			|			|			|			|			|
	on not-found			| ERR_PTR(-ENOENT)	| ERR_PTR(-ENOENT)	| -ENXIO		| dummy[1]		|
	(plain get)			|			|			|			|			|
					|			|			|			|			|
	return value			|			|			|			|			|
	on not-found			| dummy[1]		| dummy[1]		| -ENXIO		| ERR_PTR(-ENOENT)	|
	(get_optional)			|			|			|			|			|
					|			|			|			|			|
	emits an error message		|			|			|			|			|
	on error (including 		| no			| no			| yes[2]		| no			|
	not-found)			|			|			|			|			|
					|			|			|			|			|
	get_optional emits an error	|			|			|			|			|
	message on error (including	| no			| no			| no			| no			|
	not-found)			|			|			|			|			|
					|			|			|			|			|
	summary:			| returning a dummy	| returning a dummy	| doesn't emit an	| returning error code	|
	*_get_optional() differs from	| on not-found		| on not-found		| error message		| on not-found		|
	*_get by:			|			|			|			|			|


	[1] the dummy value is a valid resource descriptor, the API functions
	    are a noop for this dummy value. This dummy value is NULL for
	    all three subsystems.
	[2] no error is printed for -EPROBE_DEFER.

The inversion between clk+gpio vs. regulator is unforunate, swaping one
or the other would be good for consistency, but this isn't the topic of
this thread. Only so much: It's not agreed upon which variant is the
better one and the difference is of historical origin.

There are now different suggestions to improve the situation regarding
platform_get_irq() compared to the other functions:

a) by Sergey
   platform_get_irq_optional() is changed to return 0 on not-found.

b) by Uwe
   rename platform_get_irq_optional() to platform_get_irq_silent()

The argument pro a) is:

	platform_get_irq_optional() is aligned to clk_get() and
	gpiod_get() by returning 0 on not-found.

The argument contra a) 

	The return value 0 for platform_get_irq() is only syntactically
	nearer to the dummy value of clk_get() and gpiod_get(). A dummy
	value isn't available and probably not sensible to introduce for
	irq because most drivers have to check for the not-found
	situation anyhow to setup polling. 

The argument pro b) is:

	The relevant difference between platform_get_irq() and its
	optional variant is that the latter is silent. This is a
	different concept for the meaning of optional compared to the
	other *_get_optional().

The argument contra b) is:

	The chosen name is bad, because driver authors might wonder what
	a silent irq is.

---- end of summary
	
A possible compromise: We can have both. We rename
platform_get_irq_optional() to platform_get_irq_silent() (or
platform_get_irq_silently() if this is preferred) and once all users are
are changed (which can be done mechanically), we reintroduce a
platform_get_irq_optional() with Sergey's suggested semantic (i.e.
return 0 on not-found, no error message printking).

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

Attachment: signature.asc
Description: PGP signature