Re: [PATCH v6 1/4] lib: new helper kstrtodev_t()

[Andy Shevchenko]

On Thu, Feb 20, 2020 at 12:57 PM Uwe Kleine-KÃnig
<u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
> On Thu, Feb 20, 2020 at 12:22:36PM +0200, Andy Shevchenko wrote:
> > On Thu, Feb 20, 2020 at 9:49 AM Uwe Kleine-KÃnig
> > <u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
> > > On Wed, Feb 19, 2020 at 09:50:54PM +0200, Andy Shevchenko wrote:
> > > > On Thu, Feb 13, 2020 at 11:27 AM Uwe Kleine-KÃnig <uwe@xxxxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > This function is in the same spirit as the other kstrto* functions and
> > > > > uses the same calling convention. It expects the input string to be in
> > > > > the format %u:%u and implements stricter parsing than sscanf as it
> > > > > returns an error on trailing data (other than the usual \n).
> >
> > ...
> >
> > > > On top of that, why kstrtodev_t is so important? How many users are
> > > > already in the kernel to get an advantage out of it?
> > >
> > > Does it need to be important? It matches the other kstrto* functions and
> > > so it seemed more natural to me to put it near the other functions. I'm
> > > not aware of other potential users and surprised you seem to suggest
> > > this as a requirement.
> >
> > Yes it does. The kstrtox() are quite generic, what you are proposing
> > is rather one particular case with blurry understanding how many users
> > will be out of it.
>
> In my understanding one user is a hard requirement.

Yes. But looking at the LOCs you introduce to entire kernel in such
generic area (I wouldn't tell you anything if, for instance, you
introduced a support for hypothetical S2P bus with one host controller
driver) like lib/.

> > If you had told "look, we have 1234 users which may benefit out of
> > it", I would have given no comment against.
>
> Sure, having >1000 potential users would be a good argument pro this
> function. But having only one isn't a good contra IMHO.

For lib/ is a good argument in my opinion.

> > > > What to do with all other possible variants ("%d:%d", "%dx%d" and its
> > > > %u variant, etc)?
> > >
> > > I don't see how %d:%d is relevant, major and minor cannot be negative
> > > can they? I never saw 'x' as separator between major and minor. I
> > > considered shortly parsing %u, but given that (I think) this is an
> > > internal representation only I chose to not make it more visible than it
> > > already is.
> >
> > See above, if we are going to make it generic, perhaps better to cover
> > more possible users, right?
> > Otherwise your change provokes pile of (replaced)
> > kstrto_resolution() /* %ux:%u */
> > kstrto_range() /* %d:%d */
> > kstrto_you_name_it()
>
> Given there are respective types that this can be stored to, I don't
> object more functions of this type and don't see a good reason to not
> add such a function. And in my eyes I prefer to have such a function in
> a visible place (i.e. where all the other kstrto* functions are) to
> prevent code duplication.

You can easily satisfy above by adding a function parameter 'char
*delim', right?

> Also I don't understand yet, what you want me to do.

I have issues with kstrto() not playing with simple numbers (boolean
is a special case, but still a number at the end).
I also don't feel good with too narrow usage of the newly introduced helper

> Assume I'd be
> willing to use simple_strtoul, I'd still want to have a function that
> gives me a dev_t from a given string. Should I put this directly in my
> led-trigger driver?

I see the following possibilities:
a) put it inside the caller and forget about generic helper
b) do a generic helper, but 1/ in string_*() namespace, 2/ with a
delimiter parameter and 3/ possibility to take negative numbers

In b) case, add to the commit message how many potential _existing_
users may be converted to this.
Also it would be good to have two versions strict (only \n at the end
is allowed) and non-strict (based on the amount of users for each
group).

> > > And given that I was asked for strict
> > > parsing (i.e. not accepting 2:4:something) I'd say using simple_strto*
> > > is a step backwards. Also simple_strtoul() has "This function is obsolete.
> > > Please use kstrtoul instead." in its docstring which seems to apply to
> > > the other simple_strto*() functions, too.
> >
> > I specifically fixed a doc string to approve its use in the precisely
> > cases you have here.
>
> Can you please be a bit more constructive here and point to the change
> you talk about? I didn't find a commit in next.

https://elixir.bootlin.com/linux/v5.6-rc2/source/include/linux/kernel.h#L446

Note, there is no more word 'obsolete' there.

-- 
With Best Regards,
Andy Shevchenko