Re: [RFC PATCH] Don't reset timestamps in include/generated if not needed

[Linus Torvalds <torvalds@linux-foundation.org>]

On Tue, Mar 3, 2015 at 5:07 PM, Valdis Kletnieks
<Valdis.Kletnieks@vt.edu> wrote:
>
> Kbuild regenerates bounds.h and asm-offsets.h, resetting the timestamps
> and forcing rebuilds even if the contents haven't changed.  Add a bit of
> shell magic to only replace the file if the contents have in fact changed,
> which should speed up git bisects and similar.
> ...
> RFC because I can't wrap my head around why this wasn't done ages ago.
> If I'm missing something something obvious, please apply a cluestick. :)
>
> Lightly tested - if I rm one of those two files, it gets rebuilt. If I
> insert some whitespace, it gets replaced. If I don't touch it, the datestamp
> doesn't change.

I don't think this is wrong, but I'd really prefer to do the whole
move-if-changed thing as a separate rule, and a separate command.

IOW, could we please have something that generates the "build.h" file
by separately creating a new temporary file, and then having the
traditional kind of move-if-changed definition

   define move-if-changed
        if ! cmp -s $(1) $(2); then mv -f $(1) $(2); else rm -f $(1); fi
   endef

kind of thing that just does

   $(call move-if-changed,build.h.tmp,build.h)

Hmm? That move-if-changed can be useful for other cases too, I've seen
something like it in a lot of makefiles. I'm surprised we don't have
one already.

That said, we already have that "filechk" define that does pretty much
all of this.

So I don't like the exact patch, but I do like the effect it gives. Michal?

                    Linus