1497714 - git moz-phab fails with hm Multiple children found for commit HASH, unable to continue

Status: RESOLVED FIXED

(Conduit :: moz-phab, defect)

Description

[Bobby Holley (:bholley)]

Attachment: console output

See the attached error log

Comment 1

[Mark Côté [:mcote]]

We should investigate this before general release of git support.

Comment 2

[Piotr Zalewa [:zalun]]

I don't see anything unusual in the provided log. It seems to be a nice linear stack.
I can't replicate the issue.

Could you provide an output from the commands below?

1. `git log --graph --oneline --all`
2. `git rev-list --all --children | grep fa6de450dc09`

Comment 3

[Bobby Holley (:bholley)]

Attachment: git log --graph --oneline --all

Comment 4

[Bobby Holley (:bholley)]

Attachment: git rev-list --all --children | grep fa6de450dc09

Comment 5

[Mark Côté [:mcote]]

This is important but is going to take a bit of work when it comes to rebasing after commits are amended.  So we're going to release git support shortly as it is, with a notice that it won't yet support stacks that have branches within them.  This bug will be our next moz-phab priority.

Comment 6

[Kartikaya Gupta (email:[e-mail])]

Any update here? Just ran into this error.

Comment 7

[Aki Sasaki (not active)]

Kats: I tried the above commands and noticed that I had an extra branch from one of my commits, from a try push (branches/default/tip). I had rebased patch 2 in a patchset, but patch 1 still had an extra head pointing at branches/default/tip. I was able to successfully `moz-phab submit <patch2> <final_patch>` to get around this bug. I'm also theorizing that nuking or moving the branches/default/tip branch might help here, but I haven't tried that yet.

Comment 8

[Kartikaya Gupta (email:[e-mail])]

In my case I had checked out a new branch that was just the first commit of a longer working branch, e.g.:

git checkout -b base_branch master
// hack hack hack
git commit -am "patch 1"
// hack hack hack
git commit -am "patch 2"
git checkout -b just_part_1 HEAD~
moz-phab submit --upstream master


In this case doing `moz-phab submit master just_part_1` doesn't work either. I had to do a dummy `git commit --amend` which effectively cloned the "patch 1" commit to a new commit so that "patch 2" wasn't a descendant, and then I could submit.

Comment 9

[Piotr Zalewa [:zalun]]

Fixed in https://phabricator.services.mozilla.com/D11008

Comment 10

[Kartikaya Gupta (email:[e-mail])]

I just glanced through the patch and noticed that it attempts to check out and rebase every branch that contains the amended commits. Has this been tested in a situation with multiple git worktrees? git doesn't generally let you check out a branch if Annie worktree has the branch checked out.

i.e. something like this (I didn't try this specifically):

cd gecko
git checkout -b mypatchset master
git commit # make changes
git worktree add -b secondpatchset ../gecko2 mypatchset
cd ../gecko2
git commit # add a commit to secondpatchset on top of the first one
cd ../gecko
moz-phab submit # this might try to checkout secondpatchset and fail

Comment 11

[Kartikaya Gupta (email:[e-mail])]

s/Annie/another/ (sorry phone autocorrect)