Re: [PATCH] kbuild: compile-test global headers to ensure they are self-contained

From: Masahiro Yamada
Date: Wed Jun 26 2019 - 23:37:49 EST

Hi Sam,

On Tue, Jun 25, 2019 at 3:11 PM Sam Ravnborg <sam@xxxxxxxxxxxx> wrote:
> >
> > When all header files below include/drm are self-contained it will be a
> > single line:
> >
> > header-test-y += $(all_headers_with_subdir)
> In reality it will likely be the above, and then a list of
> header-test-n += foo.h
> For the header files that we for one or the other reason do not want to
> make self-contained.
> It would be nice to have the list of ignored files close to their home
> and not a full list in one Makefile in include/
> > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > index 3e630fcaffd1..e2f765e9d1e1 100644
> > --- a/scripts/Makefile.lib
> > +++ b/scripts/Makefile.lib
> > @@ -67,6 +67,7 @@ extra-$(CONFIG_OF_ALL_DTBS) += $(patsubst %.dtb,%.dt.yaml, $(dtb-))
> > endif
> >
> > # Test self-contained headers
> > +header-test-y := $(filter-out $(header-test-n), $(header-test-y))
> This part should include the logic to filter out duplicates too.
> I think we may do something wrong if the same header is listed twice.
> We could also extend this with a check that all files in header-test-n
> exits.
> Sam

Thanks for your comments.

Some followups:

[1] I prefer 'header-test-' to 'header-test-n'
for excluding headers.
In some places, it will be useful to
be able to write like this:
header-test-$(CONFIG_FOO) += foo.h

[2] I proposed somewhat generalized header-test-pattern-y instead of
providing both of 'all_headers' and 'all_headers_with_subdir'

BTW, "all headers" should be added with care.
scripts/Makefile.asm-generic and scripts/Makefile.headersinst
cater to removing stale headers.
But, we do not explicitly clean other headers.
We always be careful about potential matching to stale headers.

[3] I tried both 'one big single Makefile' and
'each Makefile in sub-directories'

I am slightly in favor of the former. Maybe I could be wrong,
and we may switch to the other approach.
But, I'd like to start with a single Makefile, and see how bad it is.

I sent v2:

Best Regards
Masahiro Yamada