Re: rebase --root conflicts with --committer-date-is-author-date

I think that the reason for this behaviour is that "rebase --root"
implicitly forces "rebase -i" to run (instead of the 'format-patch
piped to am'), and --committer-date-is-author-date is not supported
by the "rebase -i" codepath at all.

I see Thomas did "rebase --root" support in v1.6.2 era both in the
normal and interactive rebase codepaths, but df5df20c (rebase -i:
support --root without --onto, 2012-06-26) started forcing --root
rebase to always use the "rebase -i" codepath.

Teaching "rebase -i" to support --committer-date-is-author-date like
the normal rebase codepath is one thing, but I wonder why we force
"rebase -i" on people who merely wanted to do "--root".  The way
df5df20c (rebase -i: support --root without --onto, 2012-06-26)
explains it does not read like a valid justification, at least to
me.  It says:

    Implicitly use rebase -i for non-interactive rebase of --root
    without an --onto argument now that rebase -i can correctly do
    this.

which is "because we can", not "because it is a good thing to do for
such and such reasons".  This is an ancient change in v1.7.12 days.

Chris, do you remember if there was a reason why it was a bad idea
to teach the normal rebase codepath to handle "--root"?  I think we
would have needed to allow "am" to apply a creation patch and start
a new history on an unborn branch in order to do so, but I am not
sure if there was a valid reason why such a change to "am" would
have been a bad idea.

Thanks.

Elliot Wolk [off-list ref] writes:

============
version:

git version 2.1.4
Linux wolke-w520 3.19.3-towo.1-siduction-amd64 #1 SMP PREEMPT
siduction 3.19-9 (2015-03-26) x86_64 GNU/Linux
============
summary:

if i do `git rebase --committer-date-is-author-date --root', the
committer date is set to now, instead of set to the author date.
when rebasing onto the first commit, it works perfectly well.
============
reproduce:

set -x
cd
mkdir test
cd test
git init
echo line >> a; git add a; git commit -m 'jan1' --date '2015-01-01'
echo line >> a; git add a; git commit -m 'jan2' --date '2015-01-02'
echo line >> a; git add a; git commit -m 'jan3' --date '2015-01-03'
git log --format=format:%s-%cD-%aD%n
git rebase --committer-date-is-author-date HEAD^^
git log --format=format:%s-%cD-%aD%n
git rebase --committer-date-is-author-date --root
git log --format=format:%s-%cD-%aD%n
git rebase --committer-date-is-author-date HEAD^^
git log --format=format:%s-%cD-%aD%n
git rebase --committer-date-is-author-date --root
git log --format=format:%s-%cD-%aD%n
============
expected:

jan3-Sat, 03 Jan 2015 13:28:06 -0500-Sat, 3 Jan 2015 13:28:06 -0500
jan2-Sat, 02 Jan 2015 13:28:06 -0500-Fri, 2 Jan 2015 13:28:06 -0500
jan1-Sat, 01 Jan 2015 13:28:06 -0500-Thu, 1 Jan 2015 13:28:06 -0500
============
actual:

jan3-Sat, 11 Apr 2015 14:28:06 -0400-Sat, 3 Jan 2015 13:28:06 -0500
jan2-Sat, 11 Apr 2015 14:28:06 -0400-Fri, 2 Jan 2015 13:28:06 -0500
jan1-Sat, 11 Apr 2015 14:28:06 -0400-Thu, 1 Jan 2015 13:28:06 -0500

{full output: http://pastie.org/10086950}
============
sorry if dupe/wrong-venue/poorly-formatted/incomprehensible. i tried.

Junio C Hamano [off-list ref] wrote:

Chris, do you remember if there was a reason why it was a bad idea
to teach the normal rebase codepath to handle "--root"?  I think we
would have needed to allow "am" to apply a creation patch and start
a new history on an unborn branch in order to do so, but I am not
sure if there was a valid reason why such a change to "am" would
have been a bad idea.

Hi. It's a long time ago, but I don't remember any reason and it feels
sensible that am should be able to create an unborn branch in the same way
interactive rebase can.

I suspect I had done the necessary work for rebase -i but not for am, and
incorrectly assumed that interactive rebase was in any case a superset of
non-interactive.

Best wishes,

Chris.