Re: linux-next: Tree for December 8 (drivers/platform/x86/Kconfig:422:error: recursive dependency detected!)

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Wed, Dec 08, 2010 at 11:35:08PM +0000, David Woodhouse wrote:
> On Wed, 2010-12-08 at 15:31 -0800, Dmitry Torokhov wrote:
> > Even better tool would allow selecting the needed optios right there,
> > without the need of moving away from teh current option.
> 
> That's exactly what the Nemesis version of xconfig did, almost 15 years
> ago.
> 
> >  And yet better tool would not even ask user and enable them on its
> > own. Hey, but we have it! It's called "select".
> 
> The tools could do that with 'depends on' too. Why make a distinction?

To allow automatic/manual selection. Sometimes the only answer to
"shoudl it be enabled" is "Yes". I.e. drivers that need polling need
input-polldev support. There is no reason whatsoever to even ask user.

Same thing for platform drivers. "Don't you also want to enable option for
your keyboard and mouse to work? Go do that, you won't regret." "Well,
duh!"

> 
> > Seriously, select is dangerous. I wonder if a rule like "One can only
> > select a symbol whose dependencies are all satisfied by the current
> > symbol and/or its parents and the symbols they select or depend on"
> > would not make select safe enough. 
> 
> "One can only select a symbol which isn't otherwise presented to the
> user as a choice".

Input-polldev _is_ presented as a choice in case user or distro want to
use out of tree driver that depends on it. Does this mean we shoudl
disallow selects on it? I don't think so.

> So things like CONFIG_REED_SOLOMON are fine to be selected, because the
> user would never see it anyway.
> 
> Otherwise, just use depends. And if that's a problem for you, fix the
> damn tools.

Yep, starting with kconfig.

-- 
Dmitry