Developing components
Local development
Vite Sandbox
Running the Vite server for local development will spin up the component sandbox. Each component should have a simple working demo in the sandbox for local development and testing. These basic demos will also be used for testing components within MediaWiki.
To start the local dev server, run this command in the root of the Codex repository:
npm run dev
This will serve the sandbox page at http://localhost:5173
, and automatically update it in the browser as you make changes to the code (hot module reloading). If you need it to be served on a different port, you can use npm run dev --port=12345
.
To add a new component demo to the sandbox:
- In
packages/codex/src/demo
, add a new.vue
file containing the new demo - Import the new file into
packages/codex/src/demo/Sandbox.vue
, add it to the template, and add it to thedemoSections
array.
VitePress docs site
Aside from Vite Sandbox, you can also run the VitePress site locally to write and test a full suite of component demos. See the component demos section for more information.
Component basics
The codex
package uses Vue 3 and prefers the Composition API over the Options API.
Codex is written in TypeScript: see the Working with TypeScript section for information about code conventions, solved problems, and potential pitfalls.
Conventions
- Export all components that are ready for public consumption in
src/lib.ts
to make them available to library users. - PascalCase multi-word component names are used per the Vue.js Style Guide. Component names should be prefixed with "Cdx," e.g.
CdxButton
- Slot names should be all lowercase. Use
kebab-case
for slot names with multiple words. This is necessary to ensure support for environments that use DOM templates, including MediaWiki.
WIP components
Until a component is ready for public consumption, it is considered a "work in progress" or "WIP" component. WIP components are housed in a separate directory, packages/codex/src/components-wip
, and are included in index.ts
in that same directory. WIP components will not be included in public dist files, nor will they appear on the docs site when viewing the documentation for a specific Codex release. However, WIP components will appear on the docs site for the main branch, and when serving the docs site locally.
Adding a WIP component
To add a new WIP component:
- Add a new directory and
.vue
file to the WIP components directory, e.g.packages/codex/src/components-wip/my-new-component/MyNewComponent.vue
- Import this component into
packages/codex/src/components-wip/index.ts
Before submitting a patch for the new component, we recommend at least adding a basic demo to the sandbox and snapshot tests.
When adding examples for the VitePress demo page in the codex-docs
package, the component should be imported from the @wikimedia/codex
package as usual. This will not work in real code as long as the component is still WIP, but the codex-docs
package is set up to be able to access WIP components this way.
Transition to public component
Once the entire design spec has been implemented, the component has a full suite of interactive demos, and the component has reached the unit test threshold, it is ready for public use. To make this transition, submit a patch that:
- Moves the component directory from
components-wip
tocomponents
- Exports the component in
lib.ts
instead ofcomponents-wip/index.ts
- Updates the paths used to import component, from
../../components/foo/Foo.vue
to../foo/Foo.vue
.
Once this patch is merged, the component will be included in the dist and will appear on the docs site for the next release.
Writing styles
Styles are written in Less and are included in the single-file component at the end in a <style lang="less">
wrapped section. Codex design tokens are imported as Less variables from the @wikimedia/codex-design-tokens
package, using the default theme-wikimedia-ui
theme.
Conventions
Design tokens
See tokens organization, naming and structure for a detailed overview.
Stylesheet specific token application rules:
- If a component uses a value not represented by a Codex token yet, add a component-level Less variable in the
<style>
tag before the first selector. - Tokens should follow naming patterns established for MediaWiki.
- Codex does not use shorthand properties
background
,font
,animation
andtransition
for simpler design token scoping and code modularization reasons. Only tokens of a category type are summarized into a shorthand token, e.g.json"text-decoration": { "none": { "value": "{ text-decoration.0 }" }, "line-through": { "value": "{ text-decoration.150 }" }, "underline": { "value": "{ text-decoration.200 }" } },
Selectors and structure
- A light version of BEM is used for class naming structure:
- The root element will have the class
.cdx-component-name
- A block within that root element would have the class
.cdx-component-name__block-name
- A variation of that block would have the class
.cdx-component-name__block-name--modifier-name
- Codex actually expands the block logic structure for direct component children by a class like
.cdx-component-name__block-name__block-child-name
. While following this, there is no need to go deeper than 2 block levels in a class name; class names of further child elements can omit some of the blocks for the sake of brevity.
- The root element will have the class
- Add specificity when styling a sub-component to avoid override styles. For example, an Icon component that is inside a Label component would use the selector,
.cdx-label__icon.cdx-icon
. In Less:lesscdx-label { &__icon.cdx-icon { // ...rules... } }
- If a style or selector isn't self-explanatory, add a comment above it in Less comment style
//
. - Avoid HTML element selectors. The style rulesets should aim to be independent from specific element choices, which may change.
- Use
--has-
and--is-
prefixes for modifiers that are not tied to a specific state, e.g.--has-thumbnail
or--is-link
. - Styles specific to a component's enabled or disabled state should be contained in a selector specific to that state. This structure allows us to avoid overriding enabled styles for the disabled state.
- The pseudo-classes
&:enabled
and&:disabled
can be used when available, otherwise&--enabled
and&--disabled
classes should be added (e.g..cdx-menu-item--enabled
). These are available to contain styles for the two states and for simpler styles orientation. - The stylelint
no-descending-specificity
rule can be disabled to maintain this structure (see sample code below).
- The pseudo-classes
Less mixin parameters
- In order to distinguish clearer from CSS variables and Codex tokens, Less mixin function parameters should be prefixed with
param
(e.g.@param-size
).
Linting
- Codex uses stylelint-order to order CSS/Less rules
- Enforce relying only on specific CSS properties over shorthands for
background
,font
,animation
andtransition
. - Stylelint rules can be disabled/enabled and should be marked as CSS style comment
/* stylelint-disable-next-line rule-name */
.
Below are some sample styles for a component to demonstrate these conventions:
<style lang="less">
@import ( reference ) '@wikimedia/codex-design-tokens/theme-wikimedia-ui.less';
@import ( reference ) '../../themes/mixins/binary-input.less';
.cdx-radio {
// Common binary input styles.
.cdx-mixin-binary-input();
// Custom-styled radio that's visible to the user.
&__icon {
border-radius: @border-radius-circle;
// …
}
// HTML `<input type="radio">`.
// Based on the HTML attributes of the radio input, we can change the style of the adjacent
// `span`, which will look like a custom-styled radio.
&__input {
&:enabled {
& + .cdx-radio__icon {
border-color: @border-color-input-binary;
}
// Note: there is no focus behavior for the input in its unchecked state because you
// can't focus on it without selecting it.
&:hover + .cdx-radio__icon {
border-color: @border-color-input-binary--hover;
}
// …
}
/* stylelint-disable no-descending-specificity */
&:disabled {
// Only disabled radios should have a gray label.
& ~ .cdx-radio__label {
color: @color-disabled;
}
// …
}
/* stylelint-enable no-descending-specificity */
}
</style>
Bidirectional script support
Codex has limited support for bidirectional text. It supports pages that are entirely in a left-to-right (LTR) script, or pages that are entirely in a right-to-left (RTL) script. It does not support pages with a mix of LTR and RTL content, or pages whose directionality changes at runtime, except in some special cases. At the time of this writing, it's virtually impossible to support those use cases without the :dir()
CSS pseudo-class, which is not yet supported by most browsers.
There are tools (like postcss-rtlcss, see below) that generate bidirectional CSS using attribute selectors like [dir='ltr']
, but this technique is fragile. It breaks in confusing and ugly ways on pages that don't have a dir
attribute set, and on pages where a dir="ltr"
element is nested inside a dir="rtl"
element or vice versa.
Because of these limitations, Codex provides direction-specific stylesheets for use in MediaWiki. The codex.style.css
file contains LTR styles, and codex.style-rtl.css
contains RTL styles. The CSS files for individual components (used by MediaWiki's Codex code-splitting feature) are similarly broken down into direction-specific variants (with an -rtl
suffix). These files are described below.
Experimental Bidirectional Stylesheet
As of version 1.12.0
, the Codex package also includes a codex.style-bidi.css
file. This is an experimental stylesheet that supports client-side direction flipping via [dir]
selectors as described above. This file is not intended for usage in MediaWiki, but may be useful in single-page applications which wish to support client-side language and direction switching. The caveats above still apply.
Flipping of direction-specific styles
Styles in Codex are written for left-to-right (LTR) environments. Codex uses RTLCSS to generate flipped versions of these styles for right-to-left (RTL) environments. For example, a rule like .foo { padding-left: 4px; }
will be changed to .foo { padding-right: 4px; }
in RTL. In the build, the LTR styles are placed in codex.style.css
and the RTL styles incodex.style-rtl.css
.
In most cases, this automatic transformation should produce the correct result, but you should always test style changes in both LTR and RTL. Both the sandbox (npm run dev
) and the component demos (npm run doc:dev
) have direction switchers that allow you to switch between LTR and RTL on the fly.
In some cases, the automatic flipping transformation doesn't produce the correct result. Override directives can be used to address this. To prevent a rule from being flipped, put /* rtl:ignore */
on the line above it. To set a different value for a property in RTL, put the RTL value in a comment like /* rtl:4px */
. These two directives are the most important ones, but others exist; see the postcss-rtlcss documentation on control directives and value directives for more information.
Below is an example that demonstrates these directives:
.foo {
// This rule isn't flipped. It uses float: left; in both LTR and RTL
/* rtl:ignore */
float: left;
// This rule is flipped, because there is no rtl:ignore directive above it
// It becomes padding-right: 12px; in RTL
padding-left: 12px;
}
.bar {
// This rule uses -100% (`@size-full`) in RTL, because that value is explicitly specified.
transform: translateX( 0 ) /* rtl:translateX( -@size-full ) */;
// This rule is NOT flipped to margin-right, because an explicit RTL value is specified
// In RTL, this rule becomes margin-left: 30px; (NOT margin-right: 30px;)
margin-left: 12px /* rtl:30px */;
}
Direction-specific behavior in components
Some components need to adjust their behavior depending on the text direction. For example, components that listen for the left and right arrow keys being pressed may need to react to those key presses differently depending on the text direction.
To achieve this, components should use the useComputedDirection()
composable, which detects the direction of the surrounding context of the component at mount time. This works even on pages with mixed or nested directionality; however it does not detect changes in directionality that happen after the component is mounted.
Below is an example that demonstrates the use of the useComputedDirection()
composable function:
<template>
<!-- Set ref="rootElement" on the root element of your component -->
<div
ref="rootElement"
class="cdx-my-component"
@keydown.left.right="onKeydown"
>
<!-- ...component template goes here... -->
</div>
</template>
<script>
import { defineComponent, ref } from 'vue';
import useComputedDirection from '../../composables/useComputedDirection';
// ...other imports...
export default defineComponent( {
setup() {
const rootElement = ref(); // If using TypeScript, use ref<HTMLDivElement>();
const computedDir = useComputedDirection( rootElement );
function onKeydown( e ) {
if ( e.key === 'ArrowLeft' ) {
// Left means "next" in RTL, "previous" in LTR
navigate( computedDir.value === 'rtl' ? 'next' : 'previous' );
} else if ( e.key === 'ArrowRight' ) {
navigate( computedDir.value === 'ltr' ? 'previous' : 'next' );
}
}
function navigate( prevOrNext ) {
// ...
}
return {
rootElement,
onKeydown
};
}
} );
</script>
The Icon component also uses this composable to detect the text direction, and allows the detected direction to be overridden through the dir
prop. For more information about how bidirectionality is handled for icons in particular, see the icon documentation.
Translatable strings
Sometimes, components contain text that is always or usually the same across features. For example:
- In a Table with row selection, the "select all" checkbox always has the visually-hidden label "Select all rows".
- The "close" button in a Dialog usually has the
aria-label
"Close".
These are different from feature-specific strings, like a Table's caption or the label of a normal Button.
For strings that are always or usually the same across features, we provide two composables that do the following:
- Take in a message key.
- Look for a provided internationalization (i18n) function (like
mw.msg()
in MediaWiki). - If one is provided, get the translated string for that message key. If not, use an English fallback.
- Return the value, or if a custom value was provided for strings that can be customized, use that.
In MediaWiki core, we provide mw.msg()
as the i18n function, along with translated strings for all relevant messages, so users of Codex inside MediaWiki will get translated text for these strings by default.
Adding a new message
Every message that is used in a Codex component should also be defined in MediaWiki core. To add a new message, in MediaWiki core, add the message to both the en.json
and qqq.json
files in languages/i18n/codex
. Ensure the message key is prefixed with cdx-
and the component name.
Using the message within a component
Note that both composables return a computed ref containing the translated string.
Static strings
For strings that should always be the same and should never be customizable by the dev user, use the useI18n
composable. These strings are typically special text for assistive technology that help users understand how to use the component—we want such strings to be consistent across products.
useI18n
has 3 arguments:
Argument | Type | Description |
---|---|---|
messageKey | I18nMessageKey | The message's machine name |
defaultValue | I18nMessageValue<P> | The English fallback (as a string, or a function if there are parameters) |
params | MaybeRef<P>[] | An array of parameters for the message, if needed |
The following example comes from the Table component file; see the whole file for more context.
import useI18n from '../../composables/useI18n';
const translatedSelectAllLabel = useI18n(
// Message key.
'cdx-table-select-all-label',
// English fallback.
'Select all rows'
);
Customizable strings
For strings that are usually the same but could be customized in some cases, use the useI18nWithOverride
composable, which has 4 arguments:
Argument | Type | Description |
---|---|---|
override | Ref<string> | The component prop used as an override |
messageKey | I18nMessageKey | The message's machine name |
defaultValue | I18nMessageValue<P> | The English fallback (as a string, or a function if there are parameters) |
params | MaybeRef<P>[] | An array of parameters for the message, if needed |
Additionally, add a prop that can be used to provide a custom value for the string if desired.
The following example comes from the SearchInput component file; see the whole file for more context. SearchInput has a prop called buttonLabel
which can be used to provide custom text for the "Search" button. If this prop is set, its value is used directly, overriding the translatable message.
import useI18nWithOverride from '../../composables/useI18nWithOverride';
const translatedSearchButtonLabel = useI18nWithOverride(
// The prop that is used for customization of the string.
toRef( props, 'buttonLabel' ),
// Message key.
'cdx-search-input-search-button-label',
// English fallback.
'Search'
);
With parameters
Both composables can take in parameters for messages with dynamic parts. For example, for sortable Tables, some text is appended to the visually-hidden caption explaining that the columns are sortable. The following example comes from the Table component file; see the whole file for more context.
import useI18n from '../../composables/useI18n';
const translatedSortCaption = useI18n(
// Message key.
'cdx-table-sort-caption',
// English fallback, provided as a function to include the parameter.
( caption ) => `${ caption }, column headers with buttons are sortable.`,
// Array of parameters (in this case, just one: the caption prop).
[ toRef( props, 'caption' ) ]
);
Note that params will typically be refs to support reactivity, but they can be static values as well.
Inheriting attributes
By default, components will place any attributes bound to them on the root element of the component. Sometimes, though, we don't want this behavior. For example, for a component that contains an <input>
element, we may want to bind most of the attributes to that <input>
element rather than the root element.
Some attributes, however, should always be bound to the root element in order to provide expected results for library users. This includes class
and style
attributes.
WARNING
Binding a style
attribute to a component is highly discouraged as it could interfere with Design System consistency and negatively impact performance. Nonetheless, if one is provided, it will be bound to the root element of the component.
To help achieve the desired behavior in components like this, we have a composable called useSplitAttributes
. It provides the following:
rootClasses
: all CSS classes that should be bound to the root element, including those set via theclass
attribute on the component and those that are internal to the component, like dynamic and conditional classesrootStyle
: thestyle
attribute bound to the component, should one be providedotherAttrs
: all other attributes, which can be bound to the desired child element
Below is sample usage from the TextInput component:
<template>
<!-- Add rootClasses and rootStyle to the root element. -->
<div
class="cdx-text-input"
:class="rootClasses"
:style="rootStyle"
>
<!-- Bind otherAttrs to the input. -->
<input
v-bind="otherAttrs"
>
</div>
</template>
<script>
// Import the composable.
import useSplitAttributes from '../../composables/useSplitAttributes';
export default defineComponent( {
name: 'CdxTextInput',
// Set inheritAttrs to false.
inheritAttrs: false,
setup( props, context ) {
// Define dynamic classes internal to the component, in Vue's object
// syntax format.
const internalClasses = computed( () => {
return {
'cdx-text-input--has-start-icon': !!props.startIcon,
'cdx-text-input--has-end-icon': !!props.endIcon || props.clearable,
'cdx-text-input--clearable': isClearable.value
};
} );
// Get helpers from the composable.
const {
rootClasses,
rootStyle,
otherAttrs
} = useSplitAttributes( context.attrs, internalClasses );
}
} );
</script>