I'm using a TensorFlow encoder in my application. It works fine in my browser when the application is running but I get issues when testing that it builds:
$ npx react-scripts test --env=jsdom
FAIL src/App.test.js
● Test suite failed to run
ReferenceError: TextEncoder is not defined
16 | import TextField from '@material-ui/core/TextField';
17 | import Typography from '@material-ui/core/Typography';
> 18 | import * as mobilenet from '@tensorflow-models/mobilenet';
| ^
19 | import * as UniversalSentenceEncoder from '@tensorflow-models/universal-sentence-encoder';
20 | import * as tf from '@tensorflow/tfjs';
21 | import axios from 'axios';
at new PlatformBrowser (node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts:26:28)
at Object.<anonymous> (node_modules/@tensorflow/tfjs-core/src/platforms/platform_browser.ts:50:30)
at Object.<anonymous> (node_modules/@tensorflow/tfjs-core/src/index.ts:29:1)
at Object.<anonymous> (node_modules/@tensorflow/tfjs-converter/src/executor/graph_model.ts:18:1)
at Object.<anonymous> (node_modules/@tensorflow/tfjs-converter/src/index.ts:17:1)
at Object.<anonymous> (node_modules/@tensorflow-models/mobilenet/dist/index.js:38:14)
at Object.<anonymous> (src/components/model.js:18:1)
at Object.<anonymous> (src/App.js:8:1)
at Object.<anonymous> (src/App.test.js:3:1)
I'd like to get rid of that error. I've tried using the 'text-encoding' package but I'm not sure how get TextEncoder properly defined before the import happens.
Maybe I can set a different option for --env
?
I get the same error without --env=jsdom
. I believe I added it after getting similar types of not defined errors and it corrected an issue.
Here is my test:
import React from 'react';
import ReactDOM from 'react-dom';
import App from './App';
it('renders without crashing', () => {
const div = document.createElement('div');
ReactDOM.render(<App />, div);
ReactDOM.unmountComponentAtNode(div);
});
So setting --env=node
does not work either because: ReferenceError: document is not defined
.
jsdom
Doesn't seem to have TextEncoder
defined in global for the DOM. So you can fill it in with the node.js one.
test/custom-test-env.js:
const Environment = require('jest-environment-jsdom');
/**
* A custom environment to set the TextEncoder that is required by TensorFlow.js.
*/
module.exports = class CustomTestEnvironment extends Environment {
async setup() {
await super.setup();
if (typeof this.global.TextEncoder === 'undefined') {
const { TextEncoder } = require('util');
this.global.TextEncoder = TextEncoder;
}
}
}
npx react-scripts test --env=./test/custom-test-env.js