Re: [PATCH v3 3/5] hwmon: ltc2945: Handle error case in ltc2945_value_store

[Guenter Roeck]

On Tue, Jan 10, 2023 at 02:25:37PM -0500, Jon Cormier wrote:
> On Tue, Jan 10, 2023 at 1:22 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> >
> > On 1/10/23 10:19, Jon Cormier wrote:
> > > On Mon, Jan 9, 2023 at 7:04 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > >>
> > >> On 1/9/23 15:35, Jonathan Cormier wrote:
> > >>> ltc2945_val_to_reg errors were not being handled
> > >>> which would have resulted in register being set to
> > >>> 0 (clamped) instead of being left alone.
> > >>>
> > >>> Change reg_to_val and val_to_reg to return values
> > >>> via parameters to make it more obvious when an
> > >>> error case isn't handled. Also to allow
> > >>> the regval type to be the correct sign in prep for
> > >>> next commits.
> > >>>
> > >>
> > >> Sorry, I don't see that as reason or argument for such invasive changes.
> > >> As far as I can see, a two-liner to check the return value of val_to_reg()
> > >> should have been sufficient. Most of the rest, such as splitting
> > >> the return value into two elements, is POV and just adds additional code
> > >> and complexity for zero gain.
> > > I can do that. However, you had also mentioned changing the return
> > > type to match what the calling function was expecting, an unsigned
> > > long. But I can't do that since error codes are negative so it would
> > > be a signed long which would lose precision and seemingly defeat the
> > > point of matching the variable type the caller wants.  I could make it
> > > a signed long long but that still doesn't match.  So it seemed saner
> > > to just return the error and the value separately, that way the
> > > function declaration was explicit about the types it wanted/returned,
> > > and less room for error.  Would love to know your preferred solution.
> > >
> >
> > That is only true if the upper bit is actually ever set in that signed long.
> > Which means I'll have to verify if "would lose precision" is actually
> > a correct statement.
> I'd like to argue that is another reason to go with this change
> instead of working out the math of just how many bits are needed in
> the worst case and having to document it. And potentially getting that
> calculation wrong.  But I can if you'd like me to.

You are turning things on its head. We don't make changes like that
because of maybe. It is you who has to show that the change is
necessary, and that there is indeed a loss of precision otherwise.

Guenter