[+cc Junio, as the bug blames to him]

On Fri, Jan 10, 2014 at 02:15:40PM +0100, Francis Moreau wrote:

> In mykernel repository, I'm having 2 different behaviours with git-log
> but I don't understand why:
> Doing:
>     $ git log --oneline --cherry-pick --left-right v3.4.71-1^{}...next
> and
>     $ git log --oneline --cherry-pick --left-right v3.4.71-1...next
> give something different (where v3.4.71-1 is a tag).
> The command using ^{} looks the one that gives correct result I think.

Yeah, this looks like a bug. Here's a simple reproduction recipe:

  commit() {
    echo content >$1 &&
    git add $1 &&
    git commit -m $1

  git init repo && cd repo &&
  commit one &&
  commit two &&
  sleep 1 &&
  git tag -m foo mytag &&
  git checkout -b side HEAD^ &&
  git cherry-pick mytag &&
  commit three

The sleep seems to be necessary, to give the commit and its
cherry-picked version different commit times (presumably because it
impacts the order in which we visit them during the traversal).


  git log --oneline --decorate --cherry-pick --left-right mytag^{}...HEAD

produces the expected:

  > e36cc32 (HEAD, side) three

but running it with the tag, as:

  git log --oneline --decorate --cherry-pick --left-right mytag...HEAD


  > e36cc32 (HEAD, side) three
  > 5e96f7d two
  > db92fca (tag: mytag, master) two

Not only do we get the cherry-pick wrong (we should omit both "twos"),
but we seem to erroneously count the tagged "two" as being on the
right-hand side, which it clearly is not (and which is probably why we
don't find the match via --cherry-pick).

This worked in v1.8.4, but is broken in v1.8.5. It bisects to Junio's
895c5ba (revision: do not peel tags used in range notation, 2013-09-19),
which sounds promising.

I think what is happening is that we used to apply the SYMMETRIC_LEFT
flag directly to the commit. Now we apply it to the tag, and it does not
seem to get propagated. The patch below fixes it for me, but I have no
idea if we actually need to be setting the other flags, or just
SYMMETRIC_LEFT. I also wonder if the non-symmetric two-dot case needs to
access any pointed-to commit and propagate flags in a similar way.

diff --git a/revision.c b/revision.c
index 7010aff..1d99bfc 100644
--- a/revision.c
+++ b/revision.c
@@ -1197,6 +1197,8 @@ int handle_revision_arg(const char *arg_, struct rev_info 
*revs, int flags, unsi
                                a_flags = flags | SYMMETRIC_LEFT;
+                               a->object.flags |= a_flags;
+                               b->object.flags |= flags;
                        a_obj->flags |= a_flags;

To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to