API Design Approach

API Design Approach

We have learned a great deal regarding how Material-UI is used, and the v1 rewrite allowed us to completely rethink the component API.

API design is hard because you can make it seem simple but it's actually deceptively complex, or make it actually simple but seem complex.

@sebmarkbage

As Sebastian Markbage pointed out, no abstraction is superior to wrong abstractions. We are providing low-level components to maximize composition capabilities.

Composition

You may have noticed some inconsistency in the API regarding composing components. To provide some transparency, we have been using the following rules when designing the API:

  1. Using the children property is the idiomatic way to do composition with React.
  2. Sometimes we only need limited child composition, for instance when we don't need to allow child order permutations. In this case, providing explicit properties makes the implementation simpler and more performant; for example, the Tab takes an icon and a label property.
  3. API consistency matters.

Rules

Aside from the above composition trade-off, we enforce the following rules:

Spread

Undocumented properties supplied are spread to the root element; for instance, the className property is applied to the root.

Now, let's say you want to disable the ripples on the MenuItem. You can take advantage of the spread behavior:

<MenuItem disableRipple />

The disableRipple property will flow this way: MenuItem > ListItem > ButtonBase.

Native properties

We avoid documenting native properties supported by the DOM like className.

CSS Classes

All the components accept a classes property to customize the styles. The classes design answers two constraints: to make the classes structure as simple as possible, while sufficient to implement the Material Design specification.

  • The class applied to the root element is always called root.
  • All the default styles are grouped in a single class.
  • The classes applied to non-root elements are prefixed with the name of the element, e.g. paperWidthXs in the Dialog component.
  • The variants applied by a boolean property aren't prefixed, e.g. the rounded class applied by the rounded property.
  • The variants applied by and enum property are prefixed, e.g. the colorPrimary class applied by the color="primary" property.
  • A variant has one level of specificity. The color and variant properties are considered a variant. The lower the style specificity is, the simpler it is to override.
  • We increase the specificity for a variant modifier. We already have to do it for the pseudo-classes (:hover, :focus, etc.). It allows much more control at the cost of more boilerplate. Hopefully, it's also more intuitive.
const styles = {
  root: {
    color: green[600],
    '&$checked': {
      color: green[500],
    },
  },
  checked: {},
};

Nested components

Nested components inside a component have:

  • their own flattened properties when these are key to the top level component abstraction, for instance and id property for the Input component.
  • their own xxxProps property when users might need to tweak the internal render method's sub-components, for instance, exposing the inputProps and InputProps properties on components that use Input internally.
  • their own xxxComponent property for performing component injection.
  • their own xxxRef property when user might need to perform imperative actions, for instance, exposing a inputRef property to access the native input on the Input component. This helps answer the question "How can I access the DOM element?"

Property naming

The name of a boolean property should be chosen based on the default value. For example, the disabled attribute on an input element, if supplied, defaults to true. This choice allows the shorthand notation:

-<Input enabled={false} />
+<Input disabled />

Controlled components

Most of the controlled component are controlled via the value and the onChange properties, however, the open / onClose / onOpen combination is used for display related state.

boolean vs enum

There are two options to design the API for the variations of a component: with a boolean; or with an enum. For example, let's take a button that has different types. Each option has its pros and cons:

  • Option 1 boolean:

    type Props = {
      raised: boolean;
      fab: boolean;
    };

    This API enabled the shorthand notation: <Button>, <Button raised />, <Button fab />.

  • Option 2 enum:

    type Props = {
      variant: 'flat' | 'raised' | 'fab';
    }

    This API is more verbose: <Button>, <Button variant="contained">, <Button variant="fab">.

    However it prevents an invalid combination from being used, bounds the number of properties exposed, and can easily support new values in the future.

The Material-UI components use a combination of the two approaches according to the following rules:

  • A boolean is used when 2 degrees of freedom are required.
  • An enum is used when > 2 degrees of freedom are required, or if there is the possbility that additional degrees of freedom may be required in the future.

Going back to the previous button example; since it requires 3 degrees of freedom, we use an enum.