Re: [PATCH] kconfig: Add findconf script and helper program

[Zev Weiss]

On Wed, Jun 08, 2022 at 10:47:08PM PDT, Masahiro Yamada wrote:
> On Thu, Jun 9, 2022 at 12:49 PM Randy Dunlap <rdunlap@xxxxxxxxxxxxx> wrote:
> > On 6/8/22 20:46, Zev Weiss wrote:
> > > On Wed, Jun 08, 2022 at 07:48:44PM PDT, Randy Dunlap wrote:
> > >> Hi--
> > >>
> > >> On 6/8/22 02:54, Zev Weiss wrote:
> > >>> scripts/findconf provides menuconfig's search functionality as a
> > >>> standalone, non-interactive command, somewhat in the spirit of
> > >>> scripts/config.  It is meant to be useful for tasks like getting a
> > >>> quick overview of symbol dependencies or determining which Kconfig
> > >>> file to edit for a given symbol, without having to fire up one of the
> > >>> interactive config programs.
> > >>>
> > >>> It accepts a single command-line flag, '-v', which causes it to also
> > >>> print the help text of each matching result.
> > >>>
> > >>> Signed-off-by: Zev Weiss <zev@xxxxxxxxxxxxxxxxx>
> > >>> ---
> > >>
> > >> I can see how this could be useful.
> > >> It's a little easier to use than what I currently do:
> > >>
> > >> $ findconfig  DRM_HISI_HIBMC
> > >> ./drivers/gpu/drm/hisilicon/hibmc/Kconfig:2:config DRM_HISI_HIBMC
> > >
> > > I'm guessing 'findconfig' here is some personal shell
> > > alias/function/script?  (I can't see any references to it in the kernel
> > > source tree.)
> > >
> >
> > Yes, it's just local.
> >
> > >>
> > >> then $EDITOR that_Kconfig_file
> > >>
> > >>
> > >> In testing, I am seeing this:
> > >>
> > >> #
> > >> # using defaults found in /boot/config-5.3.18-150300.59.63-default
> > >> #
> > >> .config:421:warning: symbol value 'm' invalid for I8K
> > >> .config:2335:warning: symbol value 'm' invalid for
> > >> MTD_NAND_ECC_SW_HAMMING
> > >> .config:2484:warning: symbol value 'm' invalid for PVPANIC
> > >> .config:8671:warning: symbol value 'm' invalid for INTERCONNECT
> > >> .config:9369:warning: symbol value 'm' invalid for
> > >> CRYPTO_ARCH_HAVE_LIB_BLAKE2S
> > >> .config:9370:warning: symbol value 'm' invalid for
> > >> CRYPTO_LIB_BLAKE2S_GENERIC
> > >> .config:9653:warning: symbol value '1' invalid for KASAN_STACK
> > >>
> > >
> > > This I assume is just due to the contents of your .config file relative
> > > to the current Kconfig definitions and not a problem with anything in
> > > this patch?
> >
> > There is no .config file in the linux/ source tree at the top level.
> > I use O=build_dir for all builds.
> >
> > >
> > >> How do I specify/choose a .config file to be used?
> > >>
> > >> Oh, use KCONFIG_CONFIG=filename
> > >>
> > >
> > > Ah, I guess that'd be a nice thing to add a flag for to the wrapper
> > > script -- I'll include that in v2.
> > >
> > >>
> > >> Please update (add) usage/help text in scripts/kconfig/Makefile.
> > >>
> > >
> > > Ack, will do.
> > >
> > >
> > > Thanks for the review!
> > >
> > >
> > > Zev
> > >
>
>
>
>
>
>
>
>
> Another idea might be to add the following to
> scripts/kconfig/Makefile:
>
>
>
> @@ -77,7 +76,13 @@ PHONY += $(simple-targets)
> $(simple-targets): $(obj)/conf
>        $(Q)$< $(silent) --$@ $(Kconfig)
>
> -PHONY += savedefconfig defconfig
> +PHONY += findconfig savedefconfig defconfig
> +
> +findconfig: $(obj)/conf
> +       $(Q)$< $(silent) --$@=$(KCONFIG_FIND) $(Kconfig)
> +
> +%_findconfig: $(obj)/conf
> +       $(Q)$< $(silent) --findconfig=$* $(Kconfig)
>
> savedefconfig: $(obj)/conf
>        $(Q)$< $(silent) --$@=defconfig $(Kconfig)
>
>
>
>
>
> Instead of adding a separate program for this,
> you can modify scripts/kconfig/conf.c
>
> - add 'findconfig' to enum input_mode
> - add 'findconfig' to long_opts[]
> - add 'case findconfig' to main() function
>
>
>
> Then, you can do
>
> $ make findconfig KCONFIG_FIND=DRM_HISI_HIBMC
>
>   or
>
> $ make DRM_HISI_HIBMC_findconfig
>
>  as a shorthand.
>
>
> scripts/findconf is unneeded
> but you can put your own script in ~/bin
> if you want to save your typing even more.

Hmm, interesting idea -- it seems a bit more awkward to use though, and 
if everyone who makes much use of it is likely to be writing their own 
little ad-hoc wrapper script anyway, it seems like we might as well do 
that once and check it in?

The current approach also provides support for multi-query searches, 
which while it isn't critical is slightly more convenient than running 
it multiple times or cramming everything into a single combined regex 
(analogous to grep's '-e' flag).

That said, if you and/or others have a strong preference for doing it 
via a make target I could rearrange it to work that way instead.


Thanks,
Zev