Re: [PATCH] gpiolib: unexport gpiod_set_transitory()

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Mon, Sep 4, 2023 at 1:19 PM Andy Shevchenko
<andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:
>
> On Mon, Sep 04, 2023 at 01:14:29PM +0200, Bartosz Golaszewski wrote:
> > On Mon, Sep 4, 2023 at 12:19 PM Andy Shevchenko
> > <andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:
> > > On Sun, Sep 03, 2023 at 09:06:57PM +0200, Bartosz Golaszewski wrote:
> > > > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > > >
> > > > There are no and never have been any users of gpiod_set_transitory()
> > > > outside the core GPIOLIB code. Make it private.
> > >
> > > And rename to be gpio_desc_...()?
> > >
> > > With this done,
> > > Reviewed-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > >
> > > The rationale has been explained in the other threads with renaming matters.
> >
> > I'm not buying this explanation. Public GPIO functions don't have a
> > monopoly on the gpiod_ prefix. Eventually I'd love to unify the naming
> > convention for the three important structures that we use:
> > gpio_device, gpio_chip and gpio_desc, no matter whether they're public
> > or private as that's already clear from their placement in
> > include/linux/ or drivers/gpio/.
>
> And I would like to avoid adding confusion by mixing internal and external APIs
> under the same prefix.
>
> Personally I do not like this change, when gpiod_ is being used. So, you may
> override this, you are the maintainer, but then here is the formal NAK from me
> (as a user of these APIs internally and externally).

This is still inconsistent with the rest of the public symbols -
especially those in gpio/driver.h. My long-term plan - before making
locking great again - is to limit the usage of any gpiochip_ symbols
to GPIO providers (as the chip is not guaranteed to be valid, unlike
gpio_device) and provide a bunch of gpio_device_ or gpiodev_
interfaces for use by those who *really* need it. Now am I going to
make up two distinct prefixes for public and non-public APIs? That
will be even more confusing IMO.

Just like in C++ you don't make up special names for public vs private
methods except for some deranged coding styles that also require you
to name arguments like "in_foo" and "out_bar".

Bart