freebsd-nq/tools/tools/git/HOWTO
Ryan Stone 252b793ad4 Add some tools to simplify the use of git
Add some scripts that wraps some FreeBSD Project infrastructure
and simplifies using them with git.  The scripts are:

 - arcgit, which creates a series of reviews in Differential
 - importgit, which applies a series of git commits to svn

Differential Revision:      https://reviews.freebsd.org/D2071
2016-10-20 18:23:44 +00:00

145 lines
7.2 KiB
Plaintext

# $FreeBSD$
This directory contains tools intended to help committers use git when
interacting with standard FreeBSD project resources like Differential or svn.
I. arcgit
arcgit is a wrapper script around the arc command line tool that simplifies the
automatic creation of a series of Differential reviews from a series of git
commits. The intended workflow is:
1. Create a series of commits in git. Each commit will be a separate review, so
try to make each commit as self-contained as possible. The series of commits
should demonstrate a logical progression towards your end goal. For example,
one commit might refactor existing code to expose a new API without changing
any current functionality. A subsequent commit could then introduce your new
code that uses the new API.
It usually will not be helpful to present your code in the order in which it
was actually written and can actually be harmful. For example, if you
introduced a bug early in your development process that you fixed in a
subsequent commit, it is a waste of your reviewer's time to have them review
old code with known bugs. Instead, it would probably be best to squash the
bug fix into the commit that introduced it, so that the bug is never
presented to your reviewers in any review.
The commit headline and commit message will be imported verbatim into
Differential, so try to give each commit a meaningful commit message that
gives your reviewers the necessary context to understand your change.
2. Create your reviews bu running this command in your git repo:
$ arcgit -r C1~..C2 -R reviewer -T testplan
C1 should be the first commit that you want reviewed, and C2 should be the
last commit that you want reviewed. You may add multiple reviewers by
specifying the -R option multiple times. You can CC (AKA subscribe) people
to a review with the -C option. Note that if you subscribe a mailing list
to a review, the mailing list will be emailed for every comment or change
made to each review. Please be judicious when subscibing mailing lists to
reviews. It may be better to instead send a single email to the appropriate
list announcing all of the reviews and giving a short summary of the change
as a whole, along with a link to the individual reviews.
3. When you need to make a change and upload it to a review, use the following
process. First, check out the branch corresponding to the review (arcgit
automatically creates this branch for every review that it creates):
$ git checkout review_D1234
Next, make your change and perform whatever testing is necessary. Commit it
to your repository with this command:
$ git commit --fixup HEAD
You can upload the change to the code review by running this command in your
repository while (ensure that you are still on the review_D1234 branch):
$ arc diff --update D1234 review_D1234_base
When you run arc, it will pull up your editor and give you a chance to
change the message that will be shown in Differential for this upload. It's
recommended that you change it from the default "fixup! ...." as that does
not give your reviewers an indication of what changes were made. It's not
recommended that you give the code review fixes meaningful commit messages
directly because we will be using git's autosquash feature in the next step,
which depends on the fixup! tag being present.
Do not merge in other branches, or rebase your review branches at this point.
Any changes that are made will show up in your reviews, and that will create
extra noise that your reviewers will have to ignore. If a reviewer requests
a change that requires your commit to be based off of a later version of
head, I would suggest deferring the change from the review and creating a
new review with the change once you hit step 5.
4. Once the reviews have been approved, you need to prepare your patch series
to be committed. This involves squashing the fixes made in code review
back into the original commit that they applied to. This gives you a clean
series of commits that are ready to be commited back to svn.
First, merge each of your review branches back into your main development
branch. For example:
$ git checkout myfeature_dev
$ for branch in review_D1234 review_D1235 review_D1236; do \
git merge $branch || git mergetool -y || break; done
Next, do an interactive rebase to squash your code review fixes back into the
main commit:
$ git rebase -i -k review_D1234_base
review_D1234 should be the name of the *first* review that was created for
you in step 2. For every commit, your editor will be pulled up and you will
be given one last chance to edit your commit message. Make sure that you fill
in the "Reviewed by:" tag indicating who accepted the review. This would
be a good point to add other tags like MFC after:, Sponsored by:, etc.
The rebase will not introduce any actual changes to your code if done
properly. You can use this command to double-check that no changes were
inadvertently made here:
$ git diff myfeature_dev@{1}
5. Finally, you should get up to date with the latest changes from head:
$ git pull --rebase origin master
It is not recommended that you combine this step with the rebase done in the
previous step. The reason for this is that if you perform an interactive
rebase that changes the commit that you branch from, it becomes much more
difficult to use the reflog to confirm that the interactive rebase did not
introduce unwanted changes.
At this point, you are ready to commit your changes to head. The importgit
script can be used to import your commits directly into git.
II. importgit
importgit is a script that can take a series of commits from git and commit them
to a svn repository. The script uses the git commit messages for the svn commit
message, which allows importgit to be fully automated. This does mean that once
you start importgit, it will start commit things to svn without giving any
further chance to sanity check what it's doing.
importgit only supports importing commits that add or modify files. It does not
support importing commits that rename or delete files, to ensure that git's
rename detection heuristics do not introduce an error in the import process.
importgit also does not support importing merge commits. Only linear history
can be imported into svn.
importgit must be run from a clean subversion checkout. You should ensure that
the working tree is up-to-date with "svn up" before running importgit.
importgit will run svn directly, so make sure that your ssh-agent is running
and has your ssh key loaded already. Run importgit as follows:
$ importgit -r D1~..D2 -g /path/to/git/repo
This will import every commit between D1 and D2, including both D1 and D2. The
invocation is very similar to the invocation given to arcgit but there is an
important point to note. When you rebased your commits as you followed steps 4
and 5, the commit hashes of all of your commits changed, including C1 and C2.
You must go back and find the new commit hashes of your commits to pass to
importgit. Passing -r C1~..C2 would import your commits as they were *before*
your code review fixes were applied.