Re: [PATCH] kconfig: change .config format to use =n instead of "is not set"

[Sedat Dilek <sedat.dilek@gmail.com>]

On Sun, Feb 27, 2022 at 8:54 AM Sedat Dilek <sedat.dilek@gmail.com> wrote:
>
> On Sun, Feb 27, 2022 at 5:24 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> >
> > On Sun, Feb 27, 2022 at 6:38 AM Sedat Dilek <sedat.dilek@gmail.com> wrote:
> > >
> > > On Sat, Feb 26, 2022 at 2:34 PM Masahiro Yamada <masahiroy@kernel.org> wrote:
> > > >
> > > > The .config file uses "# CONFIG_FOO is not set" form to represent
> > > > disabled options. In the old days, it was useful because the .config
> > > > was directly included from Makefiles. For example, you can use
> > > > "ifdef CONFIG_FOO" in Makefiles to check if the option is enabled.
> > > >
> > > > Commit c955ccafc38e ("kconfig: fix .config dependencies") introduced
> > > > include/config/auto.conf, which mirrors the .config, but trims down
> > > > all disabled options.
> > > >
> > > > Since then, include/config/auto.conf defines CONFIG options during the
> > > > build. The .config is used just for storing the user's configuration.
> > > > I do not see a strong reason to use a particular pattern of comment
> > > > for disabled options.
> > > >
> > > > With this commit, Kconfig will output disable options in a more natural
> > > > form, "CONFIG_FOO=n".
> > > >
> > > > Kconfig accepts both "# CONFIG_FOO is not set" and "CONFIG_FOO=n" as a
> > > > valid input. You do not need to update arch/*/configs/*_defconfig files
> > > > for now. "git bisect" should be able to cross the commit in both ways
> > > > without any issue.
> > > >
> > >
> > > Good.
> > >
> > > Lot of people use/used the notation CONFIG_FOO=n, so did I.
> > >
> > > Thanks for keeping the "compatibility" with old usage "# CONFIG_FOO is not set".
> > >
> > > Normally, I use git diff (or scripts/diffconfig in Git tree) to
> > > compare two kernel-configs, so seeing
> > >
> > > -CONFIG_FOO=y
> > > +CONFIG_FOO=n
> > >
> > > ...might be at first view unfamiliar/unusual.
> > > With the old notation it was easier to see that Kconfig is unset.
> >
> > I agree on this point.
> >
> > "is not set" stands out much better than "=n",
> > and our eyes are accustomed to this notation for 20 years.
> >
> > However, real comments do not stand out since
> > we already (ab)use comments for disabled options.
> >
> > This is related thread
> > https://patchwork.kernel.org/project/linux-kbuild/patch/20211213100043.45645-3-arielmarcovitch@gmail.com/
> >
>
> Thanks for the link.
>
> > >
> > > Is this patch on top of kbuild-next Git?
> > >
> >
> > Yes.
>
> Let me see if I will try kbuild-next with this patch on top of
> upcoming Linux v5.17-rc6.
>

I was not able to apply your patch on top of latest kbuild-next.git:

$ git log --oneline v5.17-rc6..
3a8276b1ae7e (HEAD -> for-5.17/kbuild-next-2022022) Merge branch
'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild
into for-5.17/kbuild-next-2022022
d4c858643263 kallsyms: ignore all local labels prefixed by '.L'
a7d4f58e99dd kconfig: fix missing '# end of' for empty menu
868653f421cd kconfig: add fflush() before ferror() check
5c8166419acf kbuild: replace $(if A,A,B) with $(or A,B)
f67695c9962e kbuild: Add environment variables for userprogs flags
a5575df58004 kbuild: unify cmd_copy and cmd_shipped

$ LC_ALL=C git apply --check --verbose
../20220226_masahiroy_kconfig_change_config_format_to_use_n_instead_of_is_not_set.mbx
Checking patch scripts/kconfig/confdata.c...
error: while searching for:
       return out;
}

enum output_n { OUTPUT_N, OUTPUT_N_AS_UNSET, OUTPUT_N_NONE };

static void __print_symbol(FILE *fp, struct symbol *sym, enum output_n output_n,
                          bool escape_string)
{
       const char *val;

error: patch failed: scripts/kconfig/confdata.c:658
error: scripts/kconfig/confdata.c: patch does not apply
Checking patch scripts/kconfig/merge_config.sh...
Checking patch scripts/kconfig/streamline_config.pl...
Checking patch scripts/kconfig/tests/choice/alldef_expected_config...
Checking patch scripts/kconfig/tests/choice/allmod_expected_config...
Checking patch scripts/kconfig/tests/choice/allno_expected_config...
Checking patch scripts/kconfig/tests/choice/allyes_expected_config...
Checking patch scripts/kconfig/tests/choice/oldask1_config...
Checking patch scripts/kconfig/tests/inter_choice/expected_config...
Checking patch scripts/kconfig/tests/new_choice_with_dep/config...
Checking patch scripts/kconfig/tests/no_write_if_dep_unmet/__init__.py...
Checking patch scripts/kconfig/tests/no_write_if_dep_unmet/expected_config...

$ git log --oneline v5.17-rc6.. scripts/kconfig/confdata.c
a7d4f58e99dd kconfig: fix missing '# end of' for empty menu
868653f421cd kconfig: add fflush() before ferror() check

- Sedat -