Re: [PATCH v1 1/5] driver core: Sort headers

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

On Thu, Aug 22, 2024 at 11:30:07AM +0800, quic_zijuhu wrote:
> On 8/21/2024 11:48 PM, Andy Shevchenko wrote:
> > Sort the headers in alphabetic order in order to ease
> > the maintenance for this part.

...

> i don't think it is good idea to sort headers by alphabetic order.

I strongly disagree on this on several points:

- the header dependencies has to be resolved on each header by applying IWYU
  (Include What You Use) principle:

    in this case we don't care what is needed for each header in question

- the end developer shouldn't care about header dependencies changes as
  the project is evolving:

    it's way out of human being capacity to follow _all_ the changes in the Linux
    kernel headers

- it's much easier to maintain the inclusion block when it's sorted (to avoid
  dups, or to see in a fast manner what's already included):

    we are writing code for humans, and not for the machines (leave the
    optimisation task to the compiler in many cases)

- overall it makes the development process much easier as a whole:

    I do not believe there is a single person in the world who may tell you
    the correct order of inclusion to any, even simple, Linux kernel driver

> why ?
> 
> 1) header's dependency is not related to its file (name|path), their
> dependency are related to # includes order.

That's not true. More precisely we are working hard to make it not true (and
it's not a Plan 9 OS where as far as I know the idea was that developer knows
the exact order).

> 2) it maybe be easy to cause build error.

Yes, and again we are trying to avoid this by enforcing IWYU principle.

> 3) header's path or name maybe be related to subsystem, it is not good
> to sort one subsystem's headers before the other.

There is a grouping approach which makes this easier to get. See IIO subsystem
as a prime example for IWYU implementation in the Linux kernel.

> For header's order, my points is that:
> 
> 1) sort by their dependency.

See above. No way, it's completely impractical.

>    #include <b_header.h>
>    #include <a_header.h>
>    if
>    a_header.h:
>    #include <b_header.h>
> 
> 2) all #include <> block before all #include "" block.
> 
> 3) sort headers related to source file at the last.
> 
>    prefix_xyz.c:
> 
>    #include <>
>    .....
>    #include <prefix_xyz.h>   // it is the last if it is exposed.
> 
>    #include "internal_header.h"
>    ....
> 
> 4)
> sort relevant header together as far as possible, for example, they
> belong to the same subsystem.

-- 
With Best Regards,
Andy Shevchenko