Re: [PATCH] kbuild: take into account DT_SCHEMA_FILES changes while checking dtbs

[Dmitry Baryshkov]

On Thu, 4 Aug 2022 at 18:27, Rob Herring <robh+dt@xxxxxxxxxx> wrote:
>
> On Fri, Jul 29, 2022 at 1:46 AM Dmitry Baryshkov
> <dmitry.baryshkov@xxxxxxxxxx> wrote:
> >
> > On Fri, 29 Jul 2022 at 10:05, Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > >
> > > On Fri, Jul 29, 2022 at 3:53 PM Dmitry Baryshkov
> > > <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, 29 Jul 2022 at 08:55, Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Thu, Jul 28, 2022 at 2:36 AM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Wed, Jul 27, 2022 at 4:06 AM Dmitry Baryshkov
> > > > > > <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > It is useful to be able to recheck dtbs files against a limited set of
> > > > > > > DT schema files. This can be accomplished by using differnt
> > > > > > > DT_SCHEMA_FILES argument values while rerunning make dtbs_check. However
> > > > > > > for some reason if_changed_rule doesn't pick up the rule_dtc changes
> > > > > > > (and doesn't retrigger the build).
> > > > > > >
> > > > > > > Fix this by changing if_changed_rule to if_changed_dep and squashing DTC
> > > > > > > and dt-validate into a single new command. Then if_changed_dep triggers
> > > > > > > on DT_SCHEMA_FILES changes and reruns the build/check.
> > > > > > >
> > > > > > > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
> > > > > > > ---
> > > > > > >  scripts/Makefile.lib | 14 ++++++--------
> > > > > > >  1 file changed, 6 insertions(+), 8 deletions(-)
> > > > > > >
> > > > > > > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > > > > > > index c88b98b5dc44..3df470289382 100644
> > > > > > > --- a/scripts/Makefile.lib
> > > > > > > +++ b/scripts/Makefile.lib
> > > > > > > @@ -383,17 +383,15 @@ DT_CHECKER_FLAGS ?= $(if $(DT_SCHEMA_FILES),-l $(DT_SCHEMA_FILES),-m)
> > > > > > >  DT_BINDING_DIR := Documentation/devicetree/bindings
> > > > > > >  DT_TMP_SCHEMA := $(objtree)/$(DT_BINDING_DIR)/processed-schema.json
> > > > > > >
> > > > > > > -quiet_cmd_dtb_check =  CHECK   $@
> > > > > > > -      cmd_dtb_check =  $(DT_CHECKER) $(DT_CHECKER_FLAGS) -u $(srctree)/$(DT_BINDING_DIR) -p $(DT_TMP_SCHEMA) $@ || true
> > > > > > > +quiet_cmd_dtb =        DTC/CHECK   $@
> > > > > >
> > > > > > This is supposed to be 7 chars or less. DTCCHK or DTC_CHK perhaps. Or
> > > > > > always do just 'DTC'. I can fixup when applying.
> > > > > >
> > > > > > I'll give it a few days for other comments.
> > > > >
> > > > >
> > > > >
> > > > > When you change DT_SCHEMA_FILES, re-running dt-validate should be enough.
> > > > > You do not need to re-run dtc.
> > > > >
> > > > > I guess the strangeness comes from the fact that you are trying to do the
> > > > >  two different things in a single rule.
> > > >
> > > > The issue is that with the current rules the dt-validate isn't
> > > > re-executed on DT_SCHEMA_FILES changes. Thus comes my proposal.
> > >
> > > Correct.
> > >
> > > What I said is like this.
> > >
> > > # touch the timestamp file, %.dtb.checked
> > > $(obj)/%.dtb.checked: $(obj)/%.dtb $(DT_TMP_SCHEMA) FORCE
>
> Not really a fan of the thousands of files that creates. Maybe if it
> was turned into something useful like a list of schemas that apply to
> the dtb. IOW, a dependency list. That would speed up re-running after
> a schema change. Though if a schema change created new dependencies,
> that wouldn't work.
>
> > >         $(call if_changed_rule,dtb_check)
> > >
> > > $(obj)/%.dtb: $(src)/%.dts $(DTC) $FORCE
> > >         $(call if_changed_rule,dtc)
> > >
> > > $(obj)/%.dtbo: $(src)/%.dts $(DTC) FORCE
> > >         $(call if_changed_dep,dtc)
> > >
> > >
> > > With the dtc/check split, we can avoid unneeded regeneration of
> > > %.dtb when DT_TMP_SCHEMA or DT_SCHEMA_FILES is
> > > changed.
> > >
> > >
> > > One drawback is we track %.dtb.checked and and %.dtb separately,
> > > so something like 53182e81f47d4ea0c727c49ad23cb782173ab849
> > > may come back.
> >
> > It's up to you and Rob, but I'd really prefer a simpler solution here.
> > Regenerating dtbs sounds like a minor pain compared to hacking the
> > top-level Makefile again. What I really like is that if one has
> > CHECK_DTBS=y (for whatever reason), he can not generate dtb without
> > validation.
>
> I lean towards just rebuilding the dtbs. That's pretty quick and
> ensures we get dtc warnings with schema warnings. In the long run, I
> would like to make the schema checks not optional to run. The
> impediment to doing that is lots of warnings (but not not some
> platforms), adding a tool dependency, and validation time.

Rob, Masahiro, do we have any conclusion here? I can change my patch,
but I'd like to understand in which way I should change it.
Fixing/testing yaml changes is a bit painful w/o this change.

-- 
With best wishes
Dmitry