Re: [PATCH v2 1/2] crypto: qat - refactor included headers

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Thu, Aug 31, 2023 at 01:18:11PM +0200, Alejandro Colomar wrote:
> On 2023-08-31 05:55, Herbert Xu wrote:
> > On Tue, Aug 29, 2023 at 05:08:37PM +0300, Andy Shevchenko wrote:
> >>
> >> Do I understand correctly that you want *ideally* to have THE kernel.h
> >> as a _single_ header and that's it?
> > 
> > My rule of thumb for a .c file is that if you need more than two
> > headers directly included by kernel.h then you should just use
> > kernel.h.
> > 
> >> While I understand your motivation as a maintainer, I hate the idea of current
> >> kernel.h to be included as a silver bullet to every file because people are not
> >> capable to understand this C language part of design. The usage of the proper
> >> headers show that developer _thought_ very well about what they are doing in
> >> the driver. Neglecting this affects the quality of the code in my opinion.
> >> That's why I strongly recommend to avoid kernel.h inclusion unless it's indeed
> >> the one that provides something that is used in the driver. Even though, the
> >> rest headers also need to be included (as it wasn't done by kernel.h at any
> >> circumstances).
> > 
> > I have no qualms with fixing header files that include kernel.h
> > to include whatever it is that they need directly.  That is a
> > worthy goal and should be enforced for all new header files.
> > 
> > I just don't share your enthusiasm about doing the same for .c
> > files.
> 
> <https://include-what-you-use.org/
> 
> Maybe this is helpful, if you didn't know about it.  :)
> (I disagree with the forward declarations that are recommended there,
> though.)

Yeah, but IWYU is too radical and requires a lot of manual job done in
the kernel. Jonathan tried it at some point.

I prefer to have a balance here (not to include literally _everything_
what we are using, just generic enough).

-- 
With Best Regards,
Andy Shevchenko