|
|
This page describes the Stork project workflow. Everyone who contribute to project is expected to read it at least once and follow it. You are not special, the process applies to everyone. If you don't like it, please propose changes and discuss it with the team.
|
|
|
|
|
|
# Gitlab setup
|
|
|
- create an account if you don't have it yet.
|
|
|
- go to gitlab.isc.org, click on your user icon in right top corner and click settings from the drop menu. Then click ssh keys icon on the left panel.
|
|
|
- There are links there that explain how to generate your keys if you somehow haven't done that already. Anyway, the idea is to paste the content of your ~/.ssh/key-name.pub file there.
|
|
|
- As a completely optional step, I configured my ssh client to use separate key for gitlab, by adding this to my ~/.ssh/config file:
|
|
|
```
|
|
|
Host gitlab.isc.org
|
|
|
IdentityFile ~/.ssh/id_ed25519_gitlab
|
|
|
```
|
|
|
|
|
|
# Set up your git repository
|
|
|
|
|
|
To set up a new repository do:
|
|
|
```
|
|
|
git clone git@gitlab.isc.org:isc-projects/stork.git
|
|
|
cd stork
|
|
|
```
|
|
|
|
|
|
# Adding an issue
|
|
|
|
|
|
- If you have something new to work on, create an issue in gitlab: go to https://gitlab.isc.org/isc-projects/stork. Make sure it's not assigned to any milestone, unless you discussed this with the project manager and he is ok with adding it directly to the current milestone. We have a process to triage unassigned tickets weekly and assign them to specific milestone.
|
|
|
|
|
|
# Working on an issue
|
|
|
|
|
|
- Once you start working on an issue, assign the issue to yourself. You should also assign Doing label. There are two ways of doing that. First, you can add the label manually when browsing an issue. Alternatively, you can go to https://gitlab.isc.org/isc-projects/stork/boards and drag and drop your issue to appropriate stage.
|
|
|
|
|
|
- Open the issue page, e.g. https://gitlab.isc.org/isc-projects/stork/issues/3 and click create a MR. A good trick is to click on the triangle button to get the extended create MR menu. In particular, you can use a shorter branch name. The branch name MUST start with the issue number.
|
|
|
|
|
|
- ```git pull```. Note the branch that was just created will be there. Check it out and start working on the code. If you dislike the branch name being too long, see the trick about about using shorter name. Also, if you have bash-completion package installed (and using bash), you can type: git checkout 3, and hit tab. The full name of the branch will be completed for you.
|
|
|
|
|
|
- Commit your changes, push it on that branch.
|
|
|
|
|
|
- Once you are done, do the following: unassign yourself from the issue and MR, remove "Doing" label, add "Review" label. The doing and review labels should be put on both the issue and the MR. This indicates the MR is ready for a review. This looks like a duplication of work, but it serves different purposes. The issues are tracked using a board (see [this Stork 0.1 board](https://gitlab.isc.org/isc-projects/stork/-/boards?scope=all&utf8=%E2%9C%93&state=opened&milestone_title=Stork-0.1) for example). Once you assign a label to an issue it is shown in the appropriate column. Now, the review label on MR indicates that particular one is ready for review. The illusion of duplication disappears when there are multiple MRs assigned to a single issue.
|
|
|
|
|
|
# Reviewing an issue
|
|
|
|
|
|
- as a reviewer: go to Issues->Board page. Look for an issue in review state that has no assignee. Assign to yourself. Review. Put some snarky comments. Add some more comments. Once you're done, reassign back to the developer.
|
|
|
|
|
|
- as a developer: look for issues that are assigned to you. Do your best with addressing the comments. Push your improvements to the branch. Once done reassign back to the reviewer. DO NOT MERGE until the reviewer says the code is ready.
|
|
|
|
|
|
- as a reviewer: Once all your comments are addressed, put a note that the code is ready to merge, open the MR, edit it and click on "remove WIP status". Removal of the WIP status is a clear indication the code is ready to go.
|
|
|
|
|
|
- If you can't reach an agreement after several review rounds, you can do one of three things: ask for a third opinion (any developer will do), ask the manager to solve the problem or perhaps rethink if this is the right approach. As disappointing as it may be, sometimes abandoning the changes and starting from scratch may be the best option in the long term. Also, think on your own arguments. Are you trying to push your point of view, simply because you like it? It's an art of compromise.
|
|
|
|
|
|
# Merging code
|
|
|
|
|
|
- Stork project is currently set up in a way that allows only fast-forward merges. This is not the only possible option, but it works well for other projects (Kea, ISC DHCP) and it looks like the right way to go. It keeps the repo history much cleaner. If you complain about tons of merge conflicts, you tried to push a ticket that was too big.
|
|
|
|
|
|
- Currently we don't have any significant CI integrated with gitlab, but this will change soon. Before merging, please run unit-tests on your own.
|
|
|
|
|
|
- Using rebase button on MR page is a very nice way of rebasing your branch. If there are conflicts, here's how you can rebase the code manually:
|
|
|
```console
|
|
|
# commit everything on your branch
|
|
|
git checkout master
|
|
|
git pull
|
|
|
git checkout <your-branch>
|
|
|
git rebase master
|
|
|
# solve conflicts
|
|
|
git rebase --continue
|
|
|
git rebase --skip # if there are specific commits that you decided to skip, e.g. because they're already on master
|
|
|
```
|
|
|
|
|
|
- Once you merge your changes to master, gitlab should close the merge request and the issue.
|
|
|
|
|
|
**PROCESS CHANGE**: We previously (before Oct 2019) put three things in the Changelog: Gitlab issue number, MR number and sha of the commit-id. This uniquely identified the changes, but was annoying and it was difficult to automate (in principle it could be done, but you'd have to develop a script that would use gitlab API to retrieve the MR #). After discussion with several members of the team, I propose to put only issue #. All other details are available using GL web interface.
|
|
|
|
|
|
# GIT commit logs
|
|
|
|
|
|
In Kea we used to start every commit log with `[#Issue number,!MR number]`, but as a result of the Stork discussions, we're looking at going with a simplified approach. The proposal is to use `[#Issue number]`. This is something that can be fully automated with a pre-commit hook in gitlab, so you won't need to type it manually.
|
|
|
|
|
|
Commit logs should be descriptive. They are addressed at developers, so code familiarity by the reader is assumed, but please be considerate. "Minor changes" is not a good description. "Fixed typo in addHost()" is better.
|
|
|
|
|
|
Since we sometimes get code from places outside of our control (MRs on gitlab, PRs on github), we cannot assume the [#Issue number] information will always be available.
|
|
|
|
|
|
We don't do commit squashes in general. Sometimes the reviewer pushes fixes or we get commits from contributors (it's essential to get those preserved. Users being able to point to their code and brag about it is one of the reasons they submit patches).
|
|
|
|
|
|
If you really like to, you can squash your own commits. Nobody in the Kea team did that.
|
|
|
|
|
|
# Changelog
|
|
|
|
|
|
Changelog entries are requires for every change that is visible by a user. This covers things like new features (they may want to start using it), bug fixes (they may be affected and a fix may be the reason to upgrade), changes to build process (they may need to install new software dependency or, even when the dependency installation is automated, they should be aware that new piece of software will appear in their system) or documentation. Changelog should be modified on a branch and then merged to master. This is the same change as everything else, and should go through the usual review process.
|
|
|
|
|
|
An example changelog entry looks like this:
|
|
|
|
|
|
```
|
|
|
1662. [bug] marcin
|
|
|
Prevent deadlock in the Kea DHCP servers caused by allocating
|
|
|
memory in the system signal handler. The issue was found on
|
|
|
CentOS 7.6, but could possibly affect Kea running on any other
|
|
|
OS.
|
|
|
(Gitlab #796)
|
|
|
```
|
|
|
|
|
|
# Branch names
|
|
|
|
|
|
Branch names must start with an issue number. Yes, there has to be an issue, even if you're doing an experiment and are uncertain if the code will ever be merged. It takes 30 seconds to create one. Tomek wants to be able to dig up what you were working on when writing release notes, even if ultimately the code didn't make it to master.
|
|
|
|
|
|
# Using labels
|
|
|
|
|
|
Please try to assign some labels to your issue. If there isn't any label that's appropriate, please consider creating one. Once the project matures, it's sometimes very useful to find all issues related to certain aspect (e.g. what has changed in build procedures between 0.1 and 0.2)?
|
|
|
|
|
|
Violet labels are for priority. There are four priority labels: critical, high, medium and low.
|
|
|
|
|
|
Green labels designate type and state of the issue (Doing, Review, QA needed, etc)
|
|
|
|
|
|
Red labels are for things that should stand out. Currently there's one label: bug.
|
|
|
|
|
|
This is just a proposal. Tomek's idea was to use dark colors for important things and light colors for less important aspects. If you don't like it, we can come up with an alternative naming/coloring scheme. This is working well in Kea project.
|
|
|
|
|
|
## Working with multiple repositories (optional)
|
|
|
|
|
|
Very infrequently we'll need to access repos other than the official public one. Here's the list of optional ones, with some explanation why we could possibly need to use them.
|
|
|
|
|
|
* **trac** - We migrated to gitlab in 2018 and copied over all branches. If did some pushes to trac **after** you were supposed to do everything in trac, here's how you can still access it. ```git remote add trac ssh://tomasz@git.kea.isc.org/git/kea```
|
|
|
* **github** - We have a public clone of our repo on github. While we prefer users to send MRs on gitlab, some of thmem send PRs on github. ```git remote add github https://github.com/isc-projects/kea```
|
|
|
* **private** - When we're dealing with code changes that must not be public for a while, we use private repo. In general, it's easier to set up separate repo copy for this, but if you want to do everything on one repo, you can ```git remote add private git@gitlab.isc.org:isc-private/kea.git```
|
|
|
|
|
|
Here are a couple commands that may be useful during transition (trac->gitlab) period. However, they are useful if you happen to work with multiple repositories (such as processing pull request from github or working with internal repo).
|
|
|
|
|
|
This will pull the changes from private repo
|
|
|
```
|
|
|
git fetch private
|
|
|
```
|
|
|
|
|
|
To list your local branches with the tracking branches on remote repo:
|
|
|
```
|
|
|
git branch -vv
|
|
|
```
|
|
|
|
|
|
To list remote repos you are set to work with:
|
|
|
```
|
|
|
git remote -v
|
|
|
```
|
|
|
|
|
|
To set that your local branch should trac a remote branch, use:
|
|
|
```
|
|
|
git branch -u remote_repo/remote_branch local_branch
|
|
|
```
|
|
|
|
|
|
Fun fact: `git pull` is really a shortcut to `git pull origin`. You can do `git pull private` or `git pull github`. The pull itself does two commands: `git fetch origin` and `git merge master origin/master`.
|
|
|
|
|
|
Finally, if you can't be bothered anymore and want to get rid of this multiple repos nonsense, do this:
|
|
|
`git remote remove trac`. It will unassociate your local copy with remote repo. |
|
|
\ No newline at end of file |