Contributions to KeystoneJS in the form of issues and PRs are welcomed.
Contributions which improve the documentation and test coverage are particularly welcomed.
Keystone makes no assumptions about type of applications it powers. It achieves flexibility through small, highly composable parts that allow you to build a foundation for any type of application.
For this reason we might not add features to Keystone if they are prescriptive about:
- Data structures
- Workflows
- Access controls
- Front-end application UI
But we want your contributions! We recognise many types of applications share common features and prescriptive patterns can sometimes be helpful, even at the expense of flexibility.
If you develop custom fields, adapters, apps or any other Keystone feature, (or have an idea) join us on the Keystone Slack channel or make a pull request to KeystoneJS-Contrib and we will add it to our list of community libraries.
KeystoneJS adheres to the Contributor Covenant Code of Conduct.
KeystoneJS follows the Thinkmill Monorepo Style Guide. For more information on the reasoning behind using certain tooling, please refer to it.
Keystone uses @noviny's @changesets/cli to track package versions and publish packages.
This tool allows each PR to indicate which packages need a version bump along with a changelog snippet.
This information is then collated when performing a release to update package versions and CHANGELOG.md files.
- Make your changes (as per usual)
- Before you make a Pull Request, run the
yarn changesetcommand and answer the questions that are asked. It will want to know:- which packages you want to publish
- what version you are releasing them at
- a message to summarise the changes (this message will be written to the changelog of bumped packages)
- Before you accept the changeset, it will display all the data that will be written to the changeset. If this looks fine, agree, and a changeset will be generated in the
.changesetdirectory.
After this, a new changeset will be added which is a markdown file with YAML front matter.
-| .changeset/
-|-| UNIQUE_ID.md
The message you typed can be found in the markdown file. If you want to expand on it, you can write as much markdown as you want, which will all be added to the changelog on publish. If you want to add more packages or change the bump types of any packages, that's also fine.
While not every changeset is going to need a huge amount of detail, a good idea of what should be in a changeset is:
- WHAT the change is
- WHY the change was made
- HOW a consumer should update their code
An example, if you generate a changeset that includes adapter-mongoose as a patch, and keystone as a minor, you can merge your PR, and the next time the version-packages command is run, these will both be updated.
---
'@keystonejs/adapter-mongoose': patch
'@keystonejs/keystone': minor
---
A very useful description of the changes should be here.You can have multiple changesets in a single PR. This will give you more granular changelogs, and is encouraged.
We’re sometimes lovingly picky on the wording of our changesets because these end up in changelogs that people like you read. We want to try to get a consistent tone of voice while providing useful information to the reader.
In particular, please try to write in the past tense (e.g. "Added a new feature" rather than "Add a new feature") and write in complete sentences. This means proper capitalisation and punctuation, including full stops/periods at the end of sentences. We try to be terse when possible but if needed it's fine to write multiple lines including examples for changing APIs.
Thanks for your help with this.
This should only ever be done by a very short list of core contributors
Releasing is a two-step process. The first step updates the packages, and the second step publishes updated packages to npm.
This step is handled for us by the Changesets GitHub Action. As PRs are opened
against master, the bot will open and update a PR which generates the
appropriate CHANGELOG.md entries and package.json version bumps.
Once ready for a release, merge the bot's PR into master.
NOTE: For information on manually updating packages, see Update Packages (manual)
Once the version changes are merged back in to master, to do a manual release:
yarn fresh && \
yarn publish-changed && \
git push --tagsThe yarn publish-changed command finds packages where the version listed in the package.json is ahead of the version published on npm, and attempts to publish just those packages.
NOTE: There is no reason you should ever manually edit the version in the package.json
If you wish to do a manual release (useful for back-porting fixes), follow these steps. Otherwise, skip on to the next section for Publishing Packages.
The first step is yarn version-packages. This will find all changesets that have been created since the last release, and update the version in package.json as specified in those changesets, flattening out multiple bumps to a single package into a single version update.
The yarn version-packages command will generate a release commit, which will bump all versions, necessary dependency version changes, and update changelog.mds.
The commands to run are:
git checkout master && \
git pull && \
git branch -D temp-release-branch && \
git checkout -b temp-release-branch && \
yarn fresh && \
yarn build && \
yarn version-packages && \
yarn format && \
git add . && \
git commit -m "Version packages" && \
git push --set-upstream origin temp-release-branchOnce you have run this you will need to make a pull request to merge this back into master.
Finally, make sure you've got the latest of everything locally
git checkout master && \
git pull && \
yarnThe release process will automatically generate and update a CHANGELOG.md file, however this does not need to be the only way this file is modified. The changelogs are deliberately static assets so past changelogs can be updated or expanded upon.
In addition, content added above the last released version will automatically be appended to the next release. If your changes do not fit comfortably within the summary of a changelog, we encourage you to add a more detailed account directly into the CHANGELOG.md.
Occasionally a bug goes undetected for a few versions. When a fix is discovered, it may need to be applied to all affected versions (depending on the severity, security considerations, etc). This is called backporting.
First, find out the oldest version which was affected. This can be done using
git blame, browsing the CHANGELOG.mds, etc.
Once we know which version introduced the bug, walk forward through the
CHANGELOG.md noting all the minor and major releases made since.
Example: If a bug was introduced in
14.0.0, but not discovered until after15.1.1was released, the list of minor and major releases may look like:
14.0.x14.1.x15.0.x15.1.x
We're going to do a backport and release for the HEAD of every minor and major release, ignoring any interim patch releases.
Example: These may be the releases we'd backport to:
- ❌
14.0.0- ✅
14.0.1- ✅
14.1.0- ❌
15.0.0- ❌
15.0.1- ✅
15.0.2- ✅
15.1.0
Now, for each release we want to backport to, we follow this process:
-
Checkout the tag of the release
Let's say the package being patched is
@keystonejs/keystone, then we want to run:git checkout -b backport-keystone-14.0.1 @keystonejs/keystone@14.0.1
-
Cherry pick the commit across.
Fix any merge conflicts that might arise.
NOTE: Make sure the changeset is either regenerated or edited to accurately relfect the change to the one package you're bumping, otherwise weirdness will happen.
git cherry-pick abc123123
-
Bump package versions
yarn fresh --prefer-offline && \ yarn build && \ yarn version-packages && \ yarn format && \ git add . && \ git commit -m "Backport fix"
-
Do NOT open a PR
This change is not going to be PR'd into master. Instead we'll later push the tag which contains the commits.
To confirm everything is as expected, look at the git log:
git log -p
-
Publish the newly version bumped package
Note we can't use changesets to do this special publish as it doesn't handle backports.
(\ export PACKAGE_NAME=@keystonejs/keystone && \ export OTP_CODE= && \ cd packages/keystone && \ yarn publish --tag=backport --otp=$OTP_CODE && \ export BACKPORTED_VERSION=`npm dist-tag ls $PACKAGE_NAME | grep 'backport' | sed -e 's/backport: //'` && \ yarn tag remove $PACKAGE_NAME backport --otp=$OTP_CODE && \ git tag -a "$PACKAGE_NAME@$BACKPORTED_VERSION" -m "$PACKAGE_NAME@$BACKPORTED_VERSION" git push --tags \ )
NOTE: When prompted for "New version", just hit Enter
-
Confirm it was published
npm show <PACKAGE_NAME> versions
Some of the packages in keystone need to compiled before they're published, we use preconstruct to do this.
Preconstruct reads from the packages.json preconstruct field for configuration, in all we need to do is set the packages that we need to build. Preconstruct reads all of the packages specified in the packages and checks the main and module fields and generates compiled versions. Preconstruct under the hood uses rollup to bundle all of the modules together which means people can't expose and interact with the modules (e.g. to change internals that could change in a patch version). It also results in smaller builds because rollup is better at tree shaking than webpack. Preconstruct also uses babel and reads from the babel config in the repo to compile the code, there is one important babel plugin that Preconstruct adds which is @babel/plugin-transform-runtime which tells babel to import the helpers that it uses in generated code from a certain place rather than duplicating them in each package.
Preconstruct can generate a couple different types of modules, in keystone, we build esm and commonjs modules.
ESM bundles are built for newer bundlers like parcel, rollup and newer versions of webpack which understand ES modules and can build more optimised bundles from them than they can with commonjs.
We also build commonjs builds to run in node (for testing with jest or etc.) and for bundlers which don't understand esm. Preconstruct generates three files for commonjs, a production, development and a file to import those modules. The production one compiles out process.env.NODE_ENV !== 'production' checks which are common in front end libraries but process.env.NODE_ENV is expensive to check in node if it happens very often.
Thanks goes to these wonderful people (emoji key):
This project follows the all-contributors specification. Contributions of any kind welcome!