Skip to content

Latest commit

 

History

History
793 lines (606 loc) · 27.8 KB

developer-handbook.md

File metadata and controls

793 lines (606 loc) · 27.8 KB
Raw

Developer Handbook

Table of Contents

Getting started

Carbon is built using a collection of packages all built in the same Git repository. You might have heard this setup described as a monorepo.

As a result, we use two pieces of tooling to help us managing installing dependencies and publishing our packages. These include:

  • Yarn workspaces for handling dependencies across all packages
  • Lerna for publishing packages, tagging versions, and more

In order for you to install all the dependencies in this project, you'll need to install Yarn and run the following command in your terminal:

yarn install

This will install all of the dependencies for every package in our project. In addition, it allows us to link between packages that we are developing.

This strategy is particularly useful during development, and tooling like Lerna will pick up on when packages are linked in this way and will automatically update versions when publishing new versions of packages.

Next up, you'll most likely want to build all of the package files so that things don't fail while you are working on a package. To do this, you can run the following command:

yarn build

Afterwards, you should be good to go! For more information about how we handle dependencies, take a look at our write-up here.

Common tasks

While working on Carbon, here are some of the top-level tasks that you might want to run:

Command Usage
yarn build Uses lerna to run the build script in each package
yarn clean Resets the state of the project by removing all node_modules and running the clean script in each package
yarn doctoc Runs doctoc on all files in the doctoc directory
yarn format, yarn format:diff Format files using Prettier, check if files have been formatted
yarn sync Sync package files across the project
yarn lint Run eslint on files in the project

In addition, you can use yarn to run bin files using the yarn <bin> syntax. For example, if you wanted to use lerna to run a script in every package you could do the following:

# Access $(yarn bin)/lerna and pass `run build` to the executable
yarn lerna run build

carbon-components

Tests are written in Mocha/Chai. You can see if your code is covered by looking at carbon-components/tests/coverage/*/index.html after running test.

If your change may hit some browser quirks, use -b option, like:

gulp test:unit -b IE -b Firefox

(Other browsers tests can run with are: Safari, Chrome and ChromeHeadless)

If you are very sure that your change affects a specific set of components, you can use -f option, like:

gulp test:unit -f tests/spec/fab_spec.js

Other options for testing are:

  • -d/--debug: Stop generating code coverage report. Useful to debug your code when running test.
  • -k/--keepalive: Keep running test runner even after test ends. Test will restart running when you make changes to any test files or any files under test.
  • -v/--verbose: Let Karma emit detailed log.

Dependency management

In light of potential npm security issues [1] [2], we are addressing some of the issues with installing dependencies from a live registry by taking advantage of Yarn's offline feature. The majority of steps taken are inspired by this tweet from Lee Byron.

Continuous Integration

We specify a .yarnc file in this project that sets the path for Yarn's offline mirror to the folder .yarn/offline-mirror. This folder contains all the tarballs for the packages that the project uses. What this allows us to do is run yarn install --offline in our Continuous Integration environment so that we don't have to fetch from the live registry in our builds.

Package architecture and layout

Packages shipping Sass

If a package in elements is shipping Sass-based files, then it will follow a certain number of conventions.

The first convention is that each of these packages will have a scss folder that contains all the Sass files for the package. For example, @carbon/colors would have a folder at @carbon/colors/scss in the import path for Sass.

To include the entire package, there are two options within this scss folder: the index.scss entrypoint for modules and an entrypoint for inline support. The index.scss entrypoint would be found at @carbon/colors/scss/index.scss and would work for teams that are using tools like eyeglass or have already setup node-sass's includePaths option to include node_modules.

The other entrypoint option is for inline support. This option will work without having to use eyeglass or something like node-sass's includePaths option. Each package that ships a scss folder will include this entrypoint, and the name will reflect the package name. For example, @carbon/colors would have an entrypoint available at @carbon/colors/scss/colors.scss.

Entrypoint behavior

The entrypoints of our Sass packages will output CSS (side-effects) by default, unless there is no corresponding CSS to output. These side-effects help with quickly using a package, but sometimes an application developer will want to control behavior in order to manage side-effects. For these cases, we expose additional files that you can include in your project, most notably a common scss/mixins.scss file.

For example, in @carbon/colors we can import the carbon--colors mixin by importing @carbon/colors/scss/mixins.scss. These types of files are guaranteed to have no, or minimal, side-effects. The only side-effects that are emitted are global variable initializations as this is required behavior in newer versions of Sass.

Using these mixins.scss entrypoints allows you as an application developer to control when these side-effects are applied in your project.

Commit conventions

This project follows a structured format for writing commit messages. The main benefit of this approach is that we can use these details to automatically generate things like changelogs, in addition to clarifying what changes correspond to when looking at our Git history.

Commit message format

Parts of this section are duplicated from Angular's commit conventions.

Each commit message consists of a header, a body and a footer. The header has a specific format that includes a type, a scope and a subject:

<type>(<scope>): <subject>
<BLANK LINE>
<body>
<BLANK LINE>
<footer>

The header is mandatory and the scope of the header is optional. There are a few validation rules that we also enforce, namely that:

  • The header must always be fewer than 72 characters
  • Any line in the commit body must be fewer than 80 characters

Most of these rules are to help with integration of git with common tools.

Note: we check for this commit format using a tool called commitlint.

Type

Must be one of the following:

  • build: Changes that affect the build system or external dependencies
  • chore: Changes that do not affect the meaning of the code (white-space, formatting, missing semi-colons, etc.)
  • ci: Changes to our CI configuration files and scripts
  • docs: Documentation only changes
  • feat: A new feature
  • fix: A bug fix
  • perf: A code change that improves performance
  • refactor: A code change that neither fixes a bug nor adds a feature
  • revert: A code change that reverses a previous commit
  • test: Adding missing tests or correcting existing tests

Subject

The subject contains a succinct description of the change:

  • use the imperative, present tense: "change" not "changed" nor "changes"
  • don't capitalize the first letter
  • no dot (.) at the end

Body

Just as in the subject, use the imperative, present tense: "change" not "changed" nor "changes". The body should include the motivation for the change and contrast this with previous behavior.

Footer

The footer should contain any information about Breaking Changes.

Breaking Changes should start with the word BREAKING CHANGE: with a space or two newlines. The rest of the commit message is then used for this.

Examples

Feature (`feat`) 
// Adding new functionality to a piece of code is considered a feature.
// This can be seen as extending an existing API
-function MyComponent({ propA }) {
+function MyComponent({ propA, propB }) {
  // ...
}
Bug fix (`fix`) 
// Updating an implementation to correct a fault in the existing code is
// considered a bug fix
function add(a, b) {
-  return a - b;
+  return a + b;
}
Chore (`chore`)

Running things like formatting, or generally any project clean-up tasks, can be considered a chore that we are doing to keep things up-to-date.

Coding style

Class names

Prefix all class names with #{$prefix}-- in SCSS, which is replaced with bx-- by default, and design systems inheriting Carbon can override. This prefix prevents potential conflicts with class names from the user.

HTML

<div
  class="bx--inline-notification bx--inline-notification--error"
  role="alert"
>
  <div class="bx--inline-notification__details">...</div>
</div>

SCSS

.#{$prefix}--inline-notification {
  ...
}

.#{$prefix}--inline-notification__details {
  ...
}

Follow BEM naming convention for classes. Again, the only thing we do differently is prefix all classes with #{$prefix}--.

.#{$prefix}--block
.#{$prefix}--block__element
.#{$prefix}--block--modifier

Avoid nesting selectors, this will make it easier to maintain in the future.

// Don't do this
.#{$prefix}--inline-notification {
  .#{$prefix}--btn {
    &:hover {
      svg {
        ...
      }
    }
  }
}

// Do this instead
.#{$prefix}--inline-notification .#{$prefix}--btn {
    &:hover svg {
      ...
    }
  }
}

Sass documentation

SassDoc is used to document the Carbon Sass source. SassDoc annotations start each line with ///; do not use /// in non-SassDoc comments.

For consistency, capitalize types (used in @type, @param, @return) and descriptions (used in @param, @return, @deprecated, @example, @link).

The following annotations are used:

Required annotations

  • Description - can be one line or multiple lines

  • @access - public or private, where public items make up our public API

  • @group - typically a package or component name

  • @type - allowed on variables, (e.g. Map, Color, Number)

  • @param - allowed on functions and mixins, include the type, name, and description, with a default value if there is one (e.g. @param {Map} $breakpoints [$carbon--grid-breakpoints] - A map of breakpoints where the key is the name)

  • @return - allowed on functions, include the type and description (e.g. @return {Number} In px)

  • @alias - do not include the $ if aliasing a variable

  • @content - allowed on mixins, describe the usage of content

  • @deprecated - context around possible replacements or when the item will no longer be available

    Optional annotations

  • @example - if the usage isn't straight forward or there are multiple use cases

  • @link - if there's a related link to reference

    Examples

// Variable example

/// Primary interactive color; Primary buttons
/// @type Color
/// @access public
/// @group @carbon/themes
$interactive-01: map-get($carbon--theme, interactive-01) !default;

// Mixin example

/// Create the container for a grid. Will cause full-bleed for the grid unless
/// max-width properties are added with `make-container-max-widths`
/// @param {Map} $breakpoints [$carbon--grid-breakpoints] - A map of breakpoints where the key is the name
/// @access private
/// @group @carbon/grid
@mixin carbon--make-container($breakpoints: $carbon--grid-breakpoints) {
}

// Function example

/// Compute the type size for the given type scale step
/// @param {Number} $step - Type scale step
/// @return {Number} In px
/// @access public
/// @group @carbon/type
@function carbon--get-type-size($step) {
}

Start a new block or element?

A nested element can use a new block name as long as the styles are independent of the parent.

<div class="bx--component">
  <button class="bx--component-button">Button</button>
</div>

☝️ The #{$prefix}--component-button class implies that this button has independent styles from its parent. Generally, it's preferred to start a new block.

Red flags

Avoid names with multiple __element names:

  • ❌ .#{$prefix}--card__list__item
  • ✅ .#{$prefix}--card-item
  • ✅ .#{$prefix}--card__item

Files and folders

All components belong in src/components in their own folder.

Name files and folders using singular form; not plural.

button
  - button.hbs
  - _button.scss
  - button.js
  - button.config.js

Also note that all variants of a component can live in a single HBS, SCSS and JS file respectively. For example, while there are many button variants (primary, secondary, etc.), they're all contained in those single source files in the button folder.

Defining markups for components and their variants

There are two ways to define markups for components and their variants:

  1. Defining markup with no conditional or data interpolations
  2. Defining markup with conditionals or data interpolations

Defining markup with no conditional or data interpolations

To define markup with no conditional or data interpolation you will need to add a .hbs file to the component directory. No .config.js file in the component directory is required in this case. One thing to note is that If there is a .hbs file whose basename is exactly the same as the component name, other .hbs files has to be in componentname--variantname.hbs format.

Defining markup with conditionals or data interpolations

Defining markup with conditionals or data interpolations requires creating .config.js file, which is a JavaScript module format of Fractal configuration, in component directory. .hbs files are rendered with the data given via context property in variants[n] (below).

Supported properties in .config.js are the following:

  • default: The default variant name
  • variants - An array of objects, supporting the following properties:
    • name: The variant name
    • label: The variant name shown in dev env UI
    • notes: A short explainer the variant shown in dev env UI
    • context: The data used for rendering .hbs
    • view: The basename of the .hbs file for variant markup (Unlike default Fractal environment, this property should point to the basename of a .hbs file under demo directory or src directory, without its path)
    • preview: The basename of the .hbs file for the markup that lays out the variant markup, in "full render" mode (Unlike default Fractal environment, this property should point to the basename of a .hbs file under demo directory or src directory, without @ symbol)
    • meta: Some metadata. Carbon vanilla development environment reads the following ones specifically:
      • linkOnly: Only full-page demo is allowed
      • useIframe: Use of <iframe> for non full-page demo
      • xVersionOnly: Supports "experimental" theme only
      • xVersionNotSupported: "Experimental" theme is not supported

What .hbs file is used for rendering a variant is determined by searching for .hbs files in demo or src directory and find one whose basename matches one of the following (the priority is the following order):

  1. view property in variants[n]
  2. Variant handle, which takes a format of componentname--variantname format
  3. Component handle, which is componentname

Working on JavaScript-framework-specific styles

JavaScript-framework-specific is not recommended as we strive to create styles that are framework-neutral. However, there are some rare cases where framework-specific cannot be avoided, and some of those make sense to be in maintained by core style library here.

There are a couple ways to work on framework-specific style.

Using npm link/yarn link

This is the most straightforward way. When in the directory of your carbon-components folder, run the following command:

yarn link

You should see a success message similar to:

success Registered "carbon-components".
info You can now run `yarn link "carbon-components"` in the projects where you want to use this package and it will be used instead.

Now, go to the folder where carbon-components-angular is located and run:

yarn link carbon-components

You should see a success message similar to:

success Using linked package for "carbon-components".

The yarn link command will allow us to point the carbon-components package under node_modules to the folder on our filesystem. So, if we make a change in carbon-components and re-compile the project it will update in the Storybook environment for carbon-components-angular.

In addition, if you would like to have your changes to styles automatically compile and update Storybook you can run the following command in the carbon-components folder on your machine:

yarn gulp watch -s

This will run the watch command in gulpfile.js. As a result, whenever you make a change to the project styles it will automatically copy over into the scss folder which Storybook uses in carbon-components-angular.

Pointing NPM dependency of carbon-components right to the source code

Though above approach is the most straightforward, it involves an overhead of having to run build process at carbon-components, in addition to one at framework variant repo, upon every Sass code change.

To avoid such overhead, you can point NPM dependency of carbon-components right to the source code, though there is a caveat that our future change to directory structure, etc. may make such steps no longer work. Here are the steps:

> cd /path/to/carbon-components-angular/node_modules/carbon-components
> mv scss scss.orig
> ln -s /path/to/carbon-components/src scss

Then edits of .scss files in /path/to/carbon-components/src will be reflected to the development environment of your framework variant repository. You don't need to do anything in carbon-components side.

Maintainers

Code Patterns

Deprecating a component

Mistakes totally happen, and sometimes we'll need to figure out a strategy to remove a component from the system. When this happens, we need to make sure that users of Carbon know that:

  1. The component is going to continue to work, deprecating a component does not break any existing code
  2. That we're going to remove this component in the next major release to give them enough time to prepare

In certain cases, we'll also want to support bug fixes for deprecated components. Most of the time we will state that deprecated components will no longer receive priority fixes, but we'll still accept outside contributions.

In order to deprecate a component in our React codebase, we have the following pattern that we tend to use:

// SomeComponent.js
// React imports
import warning from 'warning';

let didWarnAboutDeprecation = false;

function SomeComponent() {
  if (__DEV__) {
    warning(
      didWarnAboutDeprecation,
      'The `SomeComponent` component has been deprecated and will be removed ' +
        'in the next major release of `carbon-components-react`'
    );
    didWarnAboutDeprecation = true;
  }
}

Note: if available, you should add a closing sentence specifying what component to use instead, or share a link for more information. This may look like:

warning(
  didWarnAboutDeprecation,
  'The `SomeComponent` component has been deprecated and will be removed ' +
    'in the next major release of `carbon-components-react`. Please use the ' +
    '`SomeOtherComponent` component instead'
);

warning(
  didWarnAboutDeprecation,
  'The `SomeComponent` component has been deprecated and will be removed ' +
    'in the next major release of `carbon-components-react`. Please visit ' +
    'ibm.biz/<some-hash> for more information'
);

Publishing

Publishing older library versions

We offer ad-hoc backwards-support for older version of the system. This work is primarly driven by external contributors who may still need these older versions for legacy code. When updates are received and merged into the codebase, the release process will be a bit different than the one described above.

For example, with carbon-components-react we have specific branches for older major versions like v5 or v6. If we wanted to publish an update to either of these major versions, this process would look like:

  • Checkout the branch locally, making sure to pull in the latest from upstream
  • Manually update package.json with the new version to publish in a branch called chore/release-vX.Y.Z and a commit message: chore(release): vX.Y.Z
  • Create a Pull Request with this new branch and commit message
  • Once this is merged into the branch, checkout locally and pull latest. Now we can publish by running npm publish ., if you want to do a dry run first you can do npm publish . --dry-run. This is helpful when dependencies may be different than in newer versions of the system

One important thing to verify is that package.json has a publishConfig field that looks like the following:

{
  "publishConfig": {
    "tag": "<VERSION>.x"
  }
}

For example, carbon-components-react v5 would look like:

{
  "publishConfig": {
    "tag": "5.x"
  }
}

This tag verifies that when we publish we do not publish to the latest tag but instead to the major-specific tag for the package.

After running npm publish . and seeing the package publish to the registry, you could create a git tag by running:

git tag -a vX.Y.Z # The commit message should match vX.Y.Z

You should then push this tag to the project by running:

git push upstream vX.Y.Z

This helps keep track of what versions have been published and snapshotting the code at that point in time.

FAQ

How do I install a dependency?

When installing a dependency, you can run yarn add <dependency-name> as normal. The only difference now is that you also will check in the corresponding tarball entry in .yarn/offline-mirror as well so that we don't have to fetch this dependency from the live registry during Continuous Integration builds.

CircleCI is failing saying that it cannot find a dependency in offline mode

Most likely this is due to Yarn mistakenly removing, or forgetting to add, a dependency to our offline mirror. Typically, running the following set of commands should reset the project back to a valid state and should bring back any missing dependencies or fetch new ones.

yarn clean
yarn cache clean
yarn

How do I fix the repo state if I cancel during a publish?

The first things Lerna will do are create a git tag and update package.json versions. If you cancel before any packages publish, then you can do the following:

# Delete the specific tag, usually something like v0.1.0
git tag -d name-of-tag
# Undo the last commit
git reset HEAD~

# Remove all staged files
git checkout -- .

You should be good to go after this!