For heavy components, some bug fixes or new features can easily destroy the original performance optimization inadvertently over time. Recently, we are refactoring the Table to troubleshoot and restore the performance loss caused by some historical updates. Here, we introduce some common troubleshooting method and frequently meet problems.
Before that, we recommend you to read the official Perf tool to choose what you need.
Render count statistics
In most cases, invalid rendering is not as dramatic as an un-optimized loop. However, in some scenarios such as large forms, tables, and lists, due to the large number of sub components, the performance impact of invalid rendering overlays is also terrible.
For example, in antd v4, in order to improve Table hover highlighting experience of rowSpan, we added an event listener for tr, and added an additional className for the selected row in td to support multiple row highlighting capability. However, because td consumes hoverStartRow and hoverEndRow data in the context, non-related rows will re-render due to changes of hoverStartRow and hoverEndRow.
Problems like this are repeated in heavy components, so we need some helper way to determine the number of renders. In the latest rc-table, we encapsulate a useRenderTimes method. It will mark the monitored rendering times on React Dev Tools through React's useDebugValue in development mode:
// Sample Code, please view real world code if needed
// Render times
const timesRef =React.useRef(0);
// Cache for prev props
const cacheProps =React.useRef(props);
const changedPropKeys =getDiff(props, cacheProps.current);// Some compare logic
Generally on the root node of the component, we will create a Context based on props and state to pass the aggregated data down. But in some cases, the actual content of the Context may not change and trigger the re-render of the child component:
In the example, although prop1 and prop2 have not changed, it is obvious that value in MyContext is a new object, causing the child component to re-render even if prop1 has not changed. So we need to Memo the Context value:
Note: You can configure eslint rules to avoid this case.
Also, refer to the example above. If we put both prop1 and prop2 in the Context, then even if prop1 does not change, prop2 changes will cause the child component to re-render. Therefore, we can split the Context into several according to the function, thereby reducing the scope of influence:
In rc-table, we split it into multiple to optimize rendering performance:
If you have used Redux, then you may be familiar with useSelector, which only rerender when the data that needs to be consumed changes. In React, there is also a related RFC(#118)(#119) about useContextSelector, which will also be implemented in React 18 in the future:
Before the API is officially launched, there are many third-party libraries implement (of course, you can also use redux directly). It is no longer necessary to consider the problem of function splitting Context through useContextSelector, which also reduces the mental burden of developers:
After optimizing in various ways, we still have to face a problem. If some rendering needs to pass through the external render method, and it happens that the method uses a closure. Then React.memo is unaware:
Due to the existence of closures, we cannot determine whether the final dom has changed before calling the render method, which is why we optimized the Table through memo in the early days of antd v4 and removed some of it over time (Actually, Table still has some scenarios where this problem needs to be solved).
Considering that Table provides shouldCellUpdate method, we plan to adjust Table rendering logic in the future. When the Parent node renders, the Table will be completely re-rendered, and when the Table is updated internally (such as horizontal scrolling position synchronization), it will hit the cache and skip.
antd Table optimization is still in progress, and we will continue to pay attention to new features of React and new ideas from the community. If you have any ideas, welcome to discuss on Github. In addition, for the suggestion of self-developed components, we recommend that after each optimization, a corresponding test case should be created, and the source issue should be noted for future retrospection. That's all. Thank you for reading.