Re: [rft, PATCH v1 4/4] kernel.h: Split out ARRAY_SZIE()

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

On Sat, Oct 29, 2022 at 01:05:43PM +0300, Alexey Dobriyan wrote:
> On Tue, Oct 25, 2022 at 11:20:10AM +0300, Andy Shevchenko wrote:
> > On Mon, Oct 24, 2022 at 09:14:29PM +0300, Alexey Dobriyan wrote:
> > > On Mon, Oct 24, 2022 at 07:05:18PM +0300, Andy Shevchenko wrote:
> > > > On Mon, Oct 24, 2022 at 06:44:50PM +0300, Alexey Dobriyan wrote:
> > > > > > include/linux/array_size.h | 13 +++++++++++++
> > > > > 
> > > > > All of this is pessimisation unless you're removing
> > > > > 
> > > > > 	+#include <linux/array_size.h>
> > > > > 
> > > > > from kernel.h which you aren't doing.
> > > > > 
> > > > > container_of.h is just as silly.
> > > > > 
> > > > > kernel.h might need _some_ cleanup (like panic and tainted stuff) which
> > > > > is rarely used but ARRAY_SIZE()?
> > > > 
> > > > Are you suggesting to slow down compilation with inclusion of tons of unneeded
> > > > stuff in the zillions of drivers?
> > > > 
> > > > Or you are talking that we need to abandon most of the headers and combine
> > > > everything into kernel.h? I think this is what is silly.
> > > 
> > > It hard to escape kernel.h so you will be including it anyway.
> > 
> > It's very important to not include it in the headers.
> > 
> > And this split helps to it a lot. We have container_of() or array_size() used
> > in macros and/or inliners that are usually appears in the headers. And if you
> > know what dependency hell means, the kernel.h brings to it a huge mess.
> > 
> > So, try to be constructive, okay?
> > 
> > > Unless you delete, say, kstrtox.h from kernel.h, it is pointless busywork.
> 
> I'm very constructive.
> 
> Parsing 1 define is faster than opening/reading/closing a file to parse
> 1 define (it's 2 defines because of header guard).
> 
> Therefore extracting 1 macro into separate file without obvious future
> growth is a pessimisation.

You are looking at it at the wrong side. The opening one file inside a single
file maybe a way to slow down the compilation, but cleaning up the _headers_
from kernel.h and similar mess is a definite win.

Ingo, for example, unwounded the thread coil for the scheduler code with a
significant win. And his work exactly shows why it's right way to go.

> And if you delete #include <array_size.h> from kernel.h and fix all
> compile failures (which you aren't doing apparently) backporters will
> hate you for life.

It would need some time to have kernel.h to be hanging as is before we can
start cleaning the mess our of it. I prefer to kill the whole header or leave
there only really _kernel_ parts.

> I've tried to delete kstrtox.h, and even allnoconfig had dozens of
> failures.

So, help us and fix them!

-- 
With Best Regards,
Andy Shevchenko