On Thu, Aug 17, 2023 at 09:12:09AM -0700, Junio C Hamano wrote:
> > diff --git a/diff.c b/diff.c
> > index 38b57b589f..1dbfdaeff0 100644
> > --- a/diff.c
> > +++ b/diff.c
> > @@ -3853,6 +3853,8 @@ static void builtin_diffstat(const char *name_a, const char *name_b,
> > && one->mode == two->mode) {
> > free_diffstat_file(file);
> > diffstat->nr--;
> > + } else {
> > + o->found_changes = 1;
> > }
> > }
> > }
>
> That is much better. In all cases where the above diffstat->nr-- is
> not reached and diffstat is kept is where we found changes, so an
> even simpler solution that fundamentally cannot go wrong would be to
> see "diffstat->nr" at the end (i.e. "are we going to show diffstat
> for *any* filepair?"). If it is non-zero, we did find a difference.
Yeah, without having really dug into the problem too far, that does
sound a lot better. I also wonder to what degree you could apply the
same strategy to other formats (I guess it depends on them removing
whitespace-only changes from a structure). From the test I posted
earlier, it does look like many of them have the same blind spots for
mode-only changes (and I suspect addition/removal of empty files is
another corner case to check).
-Peff
