On Tue, Jul 24, 2018 at 11:39 AM Mauro Carvalho Chehab
<mchehab+samsung@xxxxxxxxxx> wrote:
>
> Works for me. Do you intend to apply it directly?
Yes, I took it and it should be pushed out.
> Yeah, some time ago mailing lists got flooded with some janitorial's
> patchset adding includes (some claiming to be needed on some archs or
> under some random Kconfigs)... Compile-test ended by adding more such
> stuff (for a good reason, IMHO). I wonder if are there a better way to
> handle includes without slowing builds.
It's a nightmare to do by hand, with all the different architectures
having slightly different header file requirements.
The scheduler people did it last year (roughly Feb-2017 timeframe),
and it was painful and involved a lot of build testing. Basically some
<linux/sched.h> was split up into <linux/sched/*.h>
I wouldn't encourage people to do that again without some tooling to
actually look at "what symbols might get defined by header file
collection XYZ, what symbols might I need with any config option" kind
of logic.
But it would be lovely if somebody *could* do tooling like that.
Just having something you can run on C files that says "these headers
are completely unused under all possibly config options and
architectures" might be interesting.
Because right now, most people tend to just copy a big set of headers,
whether they need it or not. And they almost never shrink, but new
ones get added as people add uses.
Linus
