If React tries to reconcile any of the attributes you change, things will break. Here's an example:
class Application extends React.Component {
constructor() {
super();
this.state = {
classes: ["blue", "bold"]
}
}
componentDidMount() {
setTimeout(() => {
console.log("modifying state");
this.setState({
classes: this.state.classes.concat(["big"])
});
}, 2000)
}
render() {
return (
<div id="test" className={this.state.classes.join(" ")}>Hello!</div>
)
}
}
ReactDOM.render(<Application />, document.getElementById("app"), () => {
setTimeout(() => {
console.log("Adding a class manually");
const el = document.getElementById("test");
if (el.classList)
el.classList.add("grayBg");
else
el.className += ' grayBg';
}, 1000)
});
And here's the demo: https://jsbin.com/fadubo/edit?js,output
We start off with a component that has the classes blue
and bold
based on its state. After a second, we add the grayBg
class without using React. After another second, the component sets its state so that the component has the classes blue
, bold
, and big
, and the grayBg
class is lost.
Since the DOM reconciliation strategy is a black box, it's difficult to say, "Okay, my use case will work as long as React doesn't define any classes." For example, React might decide it's better to use innerHTML
to apply a large list of changes rather than setting attributes individually.
In general, if you need to do manual DOM manipulation of a React component, the best strategy is to wrap the manual operation or plugin in its own component that it can 100% control. See this post on Wrapping DOM Libs for one such example.