Re: [PATCH v8 4/4] scanf: break kunit into test cases

[Petr Mladek <pmladek@suse.com>]

On Wed 2025-03-05 10:57:47, Tamir Duberstein wrote:
> On Wed, Mar 5, 2025 at 10:55 AM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> >
> > On Wed, Mar 05, 2025 at 10:25:51AM -0500, Tamir Duberstein wrote:
> > > On Wed, Mar 5, 2025 at 10:01 AM Petr Mladek <pmladek@suse.com> wrote:
> > > > On Fri 2025-02-14 11:20:01, Tamir Duberstein wrote:
> >
> > ...
> >
> > > > >  #include <kunit/test.h>
> > > > > -#include <linux/bitops.h>
> > > > > -#include <linux/kernel.h>
> > > > >  #include <linux/module.h>
> > > > > -#include <linux/overflow.h>
> > > > > -#include <linux/printk.h>
> > > > >  #include <linux/prandom.h>
> > > > >  #include <linux/slab.h>
> > > > > -#include <linux/string.h>
> > > > > +#include <linux/sprintf.h>
> > > > >
> > > > >  #define BUF_SIZE 1024
> > > >
> > > > It would make more sense to do this clean up in the 3rd patch
> > > > where some code was replaced by the kunit macros.
> > > >
> > > > I would personally prefer to keep the explicit includes when the
> > > > related API is still used. It helps to optimize nested includes
> > > > in the header files which helps to speedup build. AFAIK, there
> > > > are people working in this optimization and they might need
> > > > to revert this change.
> > >
> > > Yeah, I don't feel strongly. I'll just restore all the includes.
> >
> > It will be blind approach. Please, try to look at them closely and include what
> > you use (IWYU principle). I don't think anybody uses kernel.h here, for
> > example.
> >
> I think I'm getting conflicting instructions here. IWYU is indeed what
> I did: bitops, kernel, overflow, printk are all unused;

I believe that the headers were added for a reason. And this patchset
keeps most of the code. This is why the change look suspicious.
And I see in the patched lib/tests/scanf_kunit.c:

  + hweight32(), BITS_PER_TYPE(), BITS_PER_LONG which looks like bitops stuff
  + is_signed_type(), type_min(), type_max() from overflow.h

So, I would keep bitops.h and overflow.h.

The printk() calls were obviously removed in the 3rd patch so printk.h
include should be removed there.

I do not see any obvious reason for kernel.h, so I would remove it
in a separate patch.

>  string is used only for sprintf, so I made that replacement.

Makes sense. But please do this in a separate patch with this
explanation. It might be done together with the kernel.h removal.

Best Regards,
Petr