Junio C Hamano wrote:
> Felipe Contreras <felipe.contreras@xxxxxxxxx> writes:
>
> > Otherwise comparing the output of commits with different dates generates
> > unnecessary diffs.
> >
> > Signed-off-by: Felipe Contreras <felipe.contreras@xxxxxxxxx>
> > ---
> > Documentation/doc-diff | 1 +
> > 1 file changed, 1 insertion(+)
>
> Ahh, it is a fix for a fallout from 28fde3a1 (doc: set actual
> revdate for manpages, 2023-04-13); when it is shown in the patch
> form like this, it is kind of obvious why we need to compensate for
> that change this way, but apparently "doc-diff" slipped everybody's
> mind back then when we were looking at the change.
Yes. doc-diff is an odd duck, because it can't be easily integrated to
the testing framework.
Sometimes a diff in the documentation is intentional, so the fact that
doc-diff generates an output from HEAD~ to HEAD is precisely what was
intended. However, sometimes it's not. Maybe a flag inside the commit
message such as GitHub's `[no ci]` might help, but it's beyond me how
could that be cleanly integrated to continous integration machinery.
For now doc-diff is meant to be run manually, therefore it's expected
that some unexpeced diffs are inevitably going to slip by, and more
relevantly: issues in doc-diff itself are going to slip by.
> Looking at the patch text of 28fde3a1, we pass GIT_VERSION and
> GIT_DATE to AsciiDoc since that version. We were already covering
> GIT_VERSION by hardcoded "omitted" string, and now we compensate for
> the other one here, which means this change and the other changes
> complement each other, and there shouldn't be a need to further
> adjustment for that change around this area. Looking good.
Yes. I think we should be passing a semi-real version instead, like
`0.0.0`, just to see how a real version would look like, but that's
orthogonal.
> > diff --git a/Documentation/doc-diff b/Documentation/doc-diff
> > index 1694300e50..554a78a12d 100755
> > --- a/Documentation/doc-diff
> > +++ b/Documentation/doc-diff
> > @@ -153,6 +153,7 @@ render_tree () {
> > make -j$parallel -C "$tmp/worktree" \
> > $makemanflags \
> > GIT_VERSION=omitted \
> > + GIT_DATE=1970-01-01 \
> > SOURCE_DATE_EPOCH=0 \
> > DESTDIR="$tmp/installed/$dname+" \
> > install-man &&
>
> I wonder what the existing SOURCE_DATE_EPOCH was trying to do there,
> though.
I also wondered the same, but again: orthogonal.
Cheers.
--
Felipe Contreras
