Re: [PATCH 1/2] i2c-algo-bit: Refactor adapter registration

[Jean Delvare <khali-PUYAD+kWke1g9hUCZPvPmw@public.gmane.org>]

On Wed, 08 Dec 2010 10:59:18 +0100, Michael Lawnick wrote:
> Jean Delvare said the following:
> > Hi Michael, Ben,
> > 
> > On Tue, 07 Dec 2010 12:59:35 +0100, Michael Lawnick wrote:
> >> Ben Dooks said the following:
> >> > On Tue, Dec 07, 2010 at 11:06:31AM +0100, Jean Delvare wrote:
> >> >> Use a function pointer to decide whether to call i2c_add_adapter or
> >> >> i2c_add_numbered_adapter. This makes the code more compact than the
> >> >> current strategy of having the common code in a separate function.
> >> > 
> >> > ok, how about changing i2c_add_numbered_adapter to take a -1 to mean
> >> > assign bus number automatically? or something similar?
> >> 
> >> IMHO better: i2c_add_adapter with optional (-1) bus parameter?
> > 
> > Which problem are you both trying to solve, please?
>
> Function pointers tend to hide information. Seeing the targeted function
> in source code makes it more clear, IMHO.

This doesn't sound like a valid argument when the provider of the
function pointer is only 20 lines away from the call site in the very
same file, sorry.

Adding a parameter to i2c_add_adapter would mean changing 105 calling
sites. You have to understand that we aren't going to do that without a
very good reason. Ben's proposal is equally invasive, as every current
call to i2c_add_adapter would have to set the id to -1 before. This
means changing 74 drivers for a marginal benefit.

If someday calls to i2c_add_numbered_adapter() outnumber calls to
i2c_add_adapter() by a factor 3, we can reconsider. But this isn't the
case today. I am not particularly happy with the current situation
myself, but it seemed like the best option when
i2c_add_numbered_adapter() was introduced, and I see no reason to
reconsider at this point in time.

-- 
Jean Delvare