- 1 Code Review in Samba - Not (Yet) Mandatory
- 2 How To Review Code
Code Review in Samba - Not (Yet) Mandatory
The Samba project does not (yet) require code reviews but encourages them. We are currently working towards establishing code reviews more broadly. Authors should try to get their patches reviewed before pushing them to master via the autobuild system.
proposed future policy
The ultimate policy we (or some of us) would like to establish for mandatory code code review is this:
- Each commit in master should have been reviewed by at least two samba team members. For this purpose, involvement in patch creation is considered as reviewing.
- The author should not push her/his own patches.
There is currently no prescribed way or tool to submit patches for review or mark patches as reviewed. The decisive point is that the patches that end up in master carry notice that they have been reviewed. We achieve this by adding certain tags to the commit messages using the command
git commit --amend
These are the tags we use:
- The Author(s) of a patch should add their signature to the commit message by adding a "Signed-off-by: " tag.
- This tag indicates that the the person it lists was involved in the creation of the patch, that it is fine with the state of the patch and submits it under the license(s) of the affected files.
- git offers the "-s" or "--signoff" switch for the sub-commands commit, cherry-pick, format-patch and am (apply-mailbox) which allows you to add the Signed-off-by tag without having to type it.
- A reviewer should add a "Reviewed-by: " tag if she was not involved in the creation of the patch.
- There is currently no git switch smilar to "--signoff", and according to git developers, there won't be. So we have to add the review tags manually.
These tags are inspired by the use in the Linux kernel. See the Linux website about submitting patches for detailed explanations. The Linux also makes use of an "Acked-by: " tag, but I suggest that we don't use it for now to keep things simple, since it does not make a lot of sense, unless we have established maintainers all over the code base and those need to Ack patches for their areas.
If the patch was created by multiple authors, they can optionally add a "Pair-programmed-with: " tag to the commit messages. This is redundant though.
function! CommitMessages() nmap S iSigned-off-by: Stefan Metzmacher <email@example.com><CR><ESC> nmap R iReviewed-by: Stefan Metzmacher <firstname.lastname@example.org><CR><ESC> iab #S Signed-off-by: Stefan Metzmacher <email@example.com> iab #R Reviewed-by: Stefan Metzmacher <firstname.lastname@example.org> iab #O Signed-off-by: iab #V Reviewed-by: iab #P Pair-Programmed-With: iab OBNOX Michael<SPACE>Adam<SPACE><email@example.com> iab VL Volker<SPACE>Lendecke<SPACE><firstname.lastname@example.org> iab METZE Stefan<SPACE>Metzmacher<SPACE><email@example.com> endf autocmd BufWinEnter COMMIT_EDITMSG,*.diff,*.patch,*.patches.txt call CommitMessages()
Where to submit for review
Patches can be sumbitted for review to the samba-technical mailing list. It is currently not required to send patches to the mailing list, so patches can of also be send to an individual reviewer directly.
How to produce patches for submission
The patches should be git-patches suitable for applying them with the "git am" command. Such patches can can either be created with the command
and attached to a mail to the mailing list or sent direclty with
If using git format-patch on a patchset (more than one commit), on can either create a series of patchfiles numbered 0001 ... 000N or use the syntax
git format-patch --stdout
to create a single mbox style file containing the whole patchset. This can then be applied with a single git am command.
Using a repository
If the author pushes his patches to a private git repository, it might be helpful to also send the gitweb url and/or git repo url, git branch name and git commit hash(es) in the patch submission email, so that potential reviewers can easily fetch the submitter's repository, check out the branch or directly cherry-pick the patch(es). This removes the need to copy around text-files.
Tracking submitted patches
We are currently evaluating the tool "patchwork" for tracking the patches that have been submitted for review to the samba-technical mailing list. A test installation is temporarily redirected from the address https://patchwork.samba.org/.
How should a reviewer act on a submitted patch?
- Apply the patchset to a local checkout of the master branch. This can be done by copying the patch file and doing "git am" or by direclty pulling the submitters git repository if available.
- Review the patches (see below for hints), do all the builds and test that seem appropriate and necessary.
- If the reviewer is OK with the patch(es) she should add the "Reviewed-by: " tag to the commit message(s) and either push the patches to master via autobuild or git format-patch the adapted patches and send them back to the author.
- If the reviewer is not OK with the patch(es), she should send comments to the author and possibly wait for the next iteration of this process.
How To Review Code
(by metze, move to own page?)
I think everybody can learn how to do good code review and I think the most important thing is that everybody reviews his/her own patches in the same way one would review patches from others!
Here are some guidelines for a successful review process:
- First read the commit message. Does that give you the impression the change is in anyway complex? If so do, a "careful_mode++".
- Read the diffstat. Does this change touches any critical areas of code? If so, do a "careful_mode++".
- First read the patch hunk by hunk and watch for Coding-Style issues. If there are strange things, do a "careful_mode++".
- Read the patch again as whole, watch for changes which could possibly change the blackbox behavior of a function.
- Depending on the "careful_mode", you should try to fully understand the change.
- Note: Here it's sometimes useful to use the -U<contextlines> feature of git show/diff, so that you see the changes in the full context of the whole function.
- If you found a change in the behavior, set "very_careful_mode = true".
- Ask yourself why a change is needed! If you don't see why it's needed, set "very_careful_mode = true".
- If you find things you don't understand, set "verfy_careful_mode = true".
- If very_careful_mode != true you may be done. This happens most of the time and most of the time careful_mode is also 0. And the costs are not too high.
- Maybe ask other people for review. Trigger discussions on the firstname.lastname@example.org mailing list.
- Read the related sources.
- Use git grep to find callers of a function.
- Do manual runtime tests.
Review your own code!
I think the most important thing is that everybody reviews his/her own patches in the same way you would review patches from others!