Re: [PATCH] build: centralize / unify asm-offsets generation

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Tue, Apr 27, 2021 at 03:40:24PM +0200, Jan Beulich wrote:
> On 27.04.2021 15:31, Roger Pau Monné wrote:
> > On Tue, Apr 27, 2021 at 02:30:35PM +0200, Jan Beulich wrote:
> >> On 27.04.2021 11:05, Roger Pau Monné wrote:
> >>> On Wed, Apr 21, 2021 at 04:09:03PM +0200, Jan Beulich wrote:
> >>>> On 20.04.2021 18:20, Roger Pau Monné wrote:
> >>>>> On Tue, Apr 20, 2021 at 05:47:49PM +0200, Jan Beulich wrote:
> >>>>>> On 20.04.2021 17:29, Roger Pau Monné wrote:
> >>>>>>> On Thu, Apr 01, 2021 at 10:33:47AM +0200, Jan Beulich wrote:
> >>>>>>>> @@ -399,7 +399,11 @@ include/xen/compile.h: include/xen/compi
> >>>>>>>>      @sed -rf tools/process-banner.sed < .banner >> $@.new
> >>>>>>>>      @mv -f $@.new $@
> >>>>>>>>  
> >>>>>>>> -include/asm-$(TARGET_ARCH)/asm-offsets.h: 
> >>>>>>>> arch/$(TARGET_ARCH)/asm-offsets.s
> >>>>>>>> +asm-offsets.s: arch/$(TARGET_ARCH)/$(TARGET_SUBARCH)/asm-offsets.c
> >>>>>>>> +    $(CC) $(filter-out -Wa$(comma)% -flto,$(c_flags)) -S -g0 -o 
> >>>>>>>> $@.new -MQ $@ $<
> >>>>>>>> +    $(call move-if-changed,$@.new,$@)
> >>>>>>>
> >>>>>>> Won't it be more natural to keep the .s file in arch/$(TARGET_ARCH)?
> >>>>>>
> >>>>>> Yes and no: Yes as far as the actual file location is concerned.
> >>>>>> No when considering where it gets generated: I generally consider
> >>>>>> it risky to generate files outside of the directory where make
> >>>>>> currently runs. There may be good reasons for certain exceptions,
> >>>>>> but personally I don't see this case as good enough a reason.
> >>>>>>
> >>>>>> Somewhat related - if doing as you suggest, which Makefile's
> >>>>>> clean: rule should clean up that file in your opinion?
> >>>>>
> >>>>> The clean rule should be in the makefile where it's generated IMO,
> >>>>> same as asm-offsets.h clean rule currently in xen/Makefile.
> >>>>>
> >>>>>> Nevertheless, if there's general agreement that keeping the file
> >>>>>> there is better, I'd make the change and simply ignore my unhappy
> >>>>>> feelings about it.
> >>>>>
> >>>>> I don't have a strong opinion, it just feels weird to have this IMO
> >>>>> stray asm-offsets.s outside of it's arch directory, taking into
> >>>>> account that we have asm-offsets.h generated from xen/Makefile into an
> >>>>> arch specific directory already as a precedent in that makefile.
> >>>>
> >>>> Well, asm-offsets.h generation doesn't involve the compiler, hence
> >>>> no .*.d files get generated and want including later. For
> >>>> asm-offsets.s to have dependencies properly honored, if we
> >>>> generated it in xen/arch/<arch>, .asm-offsets.d would also end up
> >>>> there, and hence including of it would need separately taking care
> >>>> of.
> >>>
> >>> I'm not sure I understand what do you mean with inclusion need taking
> >>> care of separately. Isn't it expected for .d file to reside in the
> >>> same directory as the output,
> >>
> >> Yes, ...
> >>
> >>> and hence picked up automatically?
> >>
> >> ... and hence they _wouldn't_ be picked up automatically while
> >> building in xen/ if the .s file got created in xen/arch/<arch>/.
> > 
> > Hm, so that would prevent re-building the target when the dependencies
> > changed as the .d file being in the arch directory would attempt the
> > rebuild from there instead of the top level xen/?
> 
> No, in the arch directory nothing should happen at all, as there's
> no rule to rebuild asm-offsets.s. And if we built it in the subarch
> directory (where asm-offsets.c lives), the wrong rule would kick in
> (the general one compiling C to assembly).
> 
> > I guess the alternative would be to force a rebuild of the target
> > every time, in order to not depend on the .d logic?
> 
> Simply rebuilding always is not going to be good: There should be
> no re-building at all when actually just installing Xen. Hence
> while move-if-changed would be able to suppress the bulk of the
> fallout, this would still be too much rebuilding for my taste in
> that specific case.
> 
> The option I've been hinting at was to explicitly include the .d
> files from the arch dir. But I don't really like this either ...

It's hard to tell whether I would prefer that option without seeing
it. In any case, the change is an improvement overall as the logic
gets shared between architectures, so I don't plan to hold it just
because of the placement nit:

Acked-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>

Thanks, Roger.