Re: [PATCH v25 01/16] dt: bindings: Add multicolor class dt bindings documention

From: Rob Herring
Date: Thu Jun 04 2020 - 18:35:28 EST


On Thu, Jun 4, 2020 at 6:51 AM Pavel Machek <pavel@xxxxxx> wrote:
>
> On Tue 2020-06-02 15:44:32, Rob Herring wrote:
> > On Tue, Jun 2, 2020 at 2:04 PM Pavel Machek <pavel@xxxxxx> wrote:
> > >
> > > On Wed 2020-05-27 08:35:06, Rob Herring wrote:
> > > > On Wed, May 27, 2020 at 7:39 AM Pavel Machek <pavel@xxxxxx> wrote:
> > > > >
> > > > > Hi!
> > > > >
> > > > > Thanks for reviews!
> > > > >
> > > > > > > +additionalProperties: false
> > > > > > > +...
> > > > > > > diff --git a/drivers/leds/led-core.c b/drivers/leds/led-core.c
> > > > > >
> > > > > > This isn't a binding file. Belongs in another patch.
> > > > >
> > > > > These constants are directly related to the binding. It makes sense to
> > > > > go in one patch...
> > > >
> > > > Yes, the header does go in this patch, but kernel subsystem files do not.
> > > >
> > > > Part of the reason for separating is we generate a DT only repository
> > > > which filters out all the kernel code. Ideally this is just filtering
> > > > out commits and the commit messages still make sens
> > >
> > > Well, but the patch can't be split like that. Otherwise we risk null
> > > pointer dereferences when one part is applied but not the second one.
> >
> > There's no risk because you are supposed to apply both patches. I
> > don't apply binding patches that are a part of a series like this.
>
> Yes, this is always guaranteed to happen, because "git bisect"
> understand patch series. Oh, wait.

What!? If the binding patch with the header comes first, how would
bisect build the driver change without the header?

> Patches are supposed to be correct on their own. If your repository
> filtering can not handle that, you need to fix that...

I'm just asking you to follow the process that *everyone* else is
following and works. It's not really about the repository filtering.
That doesn't care. A binding ABI is defined by the schema and any
defines it has. That is the logical unit that stands on its own.

Rob