Re: [RFC PATCH v4] Implement --first-parent for git rev-list --bisect

[Tiago Botelho <tiagonbotelho@gmail.com>]

On 28 May 2018 at 15:25, Junio C Hamano <gitster@pobox.com> wrote:

> Tiago Botelho <tiagonbotelho@gmail.com> writes:

> > This will enable users to implement bisecting on first parents
> > which can be useful for when the commits from a feature branch
> > that we want to merge are not always tested.
> >
> > Signed-off-by: Tiago Botelho <tiagonbotelho@hotmail.com>
> > ---
> >
> > This patch adds all Junio's suggestions, namely do_find_bisection()
being
> > broken when assigning q's weight to p if in first parent mode and q
being
> > not UNINTERESTING and its weight still not being known.
> >
> > The graph displayed in the unit tests was also changed from being
top-bottom
> > to be left-right in order to keep it consistent with graphs in other
tests.
> >
> >  bisect.c                   | 45
++++++++++++++++++++++++++++++---------------
> >  bisect.h                   |  3 ++-
> >  builtin/rev-list.c         |  3 +++
> >  revision.c                 |  3 ---
> >  t/t6002-rev-list-bisect.sh | 37 +++++++++++++++++++++++++++++++++++++
> >  5 files changed, 72 insertions(+), 19 deletions(-)

> > diff --git a/t/t6002-rev-list-bisect.sh b/t/t6002-rev-list-bisect.sh
> > index a66140803..774d9a4fd 100755
> > --- a/t/t6002-rev-list-bisect.sh
> > +++ b/t/t6002-rev-list-bisect.sh
> > @@ -263,4 +263,41 @@ test_expect_success 'rev-parse --bisect can
default to good/bad refs' '
> > ...
> > +test_output_expect_success "--bisect-all --first-parent" 'git rev-list
--bisect-all --first-parent FX ^A' <<EOF
> > +$(git rev-parse EX) (dist=1)
> > +$(git rev-parse D) (dist=1)
> > +$(git rev-parse FX) (dist=0)
> > +EOF
> > +
> >  test_done

> Running this test number of times gives me spurious errors.  Is the
> order of these output lines unstable?  How do we "sort" these
> bisect-all results?  If we are not sorting and the output depends on
> happenstance, then probably we would need to compare the expected
> and actual output after sorting.  Or if the output depends on
> something under our control (e.g. they are related to topology and
> relative commit timestamp), we probably should try to control that
> "something" tighter so that we can rely on the order of the lines in
> the "expect" file.

The reason why the tests were failing was because the above "old" tests
did not make use of test_commit which in turn would make the sha of each
commit be different and as a result give unexpected outputs at times.
If I move them to the top of that file the tests will pass every time,
would that
be ok?

> It also appears that we have "--bisect and --first-parent do not
> work well together" in t6000, which also needs to be updated.  I
> needed the following squashed into this patch to make "make test"
> pass.

> diff --git a/t/t6000-rev-list-misc.sh b/t/t6000-rev-list-misc.sh
> index 969e4e9e52..981198ae6e 100755
> --- a/t/t6000-rev-list-misc.sh
> +++ b/t/t6000-rev-list-misc.sh
> @@ -96,8 +96,8 @@ test_expect_success 'rev-list can show index objects' '
>          test_cmp expect actual
>   '

> -test_expect_success '--bisect and --first-parent can not be combined' '
> -       test_must_fail git rev-list --bisect --first-parent HEAD
> +test_expect_success '--bisect and --first-parent can now be combined' '
> +       git rev-list --bisect --first-parent HEAD
>   '

>   test_expect_success '--header shows a NUL after each commit' '