How does fetch detect if a request using a relative path is coming from a browser?

I am making a request from a react app using a relative path "/api/test". (an NGINX controller routes traffic to either the api or the react client, so they have the same base url)

I found in the msw documentation that relative paths are supported.

// Given your application runs on "http://localhost:8080",
// this request handler URL gets resolved to "http://localhost:8080/invoices"
rest.get('/invoices', invoicesResolver)

Note that relative URL are resolved against the current location (location.origin).

If I console.log(window.location.origin), it prints http://localhost:3000. I think that the origin is being set because of import '@testing-library/jest-dom'.

However, if I make a fetch call during testing (msw is mocking api calls), I get a TypeError: Invalid URL: /api/test.

fetch('/api/test')
    .then((response) => response.json())
    .then((data) => console.log(data))

Vitest caught 1 unhandled error during the test run.
This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ Unhandled Rejection ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
TypeError: Failed to parse URL from /api/test
 ❯ new Request node:internal/deps/undici/undici:9474:19
 ❯ FetchInterceptor.<anonymous> node_modules/@mswjs/interceptors/src/interceptors/fetch/index.ts:42:23
 ❯ step node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:59:23
 ❯ Object.next node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:40:53
 ❯ node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:34:71
 ❯ __awaiter node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:30:12
 ❯ globalThis.fetch node_modules/@mswjs/interceptors/src/interceptors/fetch/index.ts:41:42
 ❯ App src/App.tsx:9:3
      7|
      8| function App() {
      9|   fetch('/api/test')
       |   ^
     10|     .then((response) => response.json())
     11|     .then((data) => console.log(data))
 ❯ renderWithHooks node_modules/react-dom/cjs/react-dom.development.js:16305:18

This error originated in "src/App.test.tsx" test file. It doesn't mean the error was thrown inside the file itself, but while it was running.
Caused by: TypeError: Invalid URL: /api/test
 ❯ new URLImpl node_modules/whatwg-url/lib/URL-impl.js:21:13
 ❯ Object.exports.setup node_modules/whatwg-url/lib/URL.js:54:12
 ❯ new URL node_modules/whatwg-url/lib/URL.js:115:22
 ❯ new Request node:internal/deps/undici/undici:9472:25
 ❯ FetchInterceptor.<anonymous> node_modules/@mswjs/interceptors/src/interceptors/fetch/index.ts:42:23
 ❯ step node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:59:23
 ❯ Object.next node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:40:53
 ❯ node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:34:71
 ❯ __awaiter node_modules/@mswjs/interceptors/lib/interceptors/fetch/index.js:30:12

When I run the code in the browser, there are no issues. Given that document.baseURI and window.location.origin are set, how does fetch "know" it is not running in a browser? Is it possible to avoid this error by setting another field in window?

Rather than try to grok environment and include special behavior...

The global fetch function in one runtime might not implement the same relative URL resolution behavior as another — especially given implicit global state (such as globalThis.window), and the fact that some environments mock the function or replace it with a custom proxy.

In order to implement deterministic resolution, you can provide a fully-qualified address at the time of invocation — this will guarantee reliable behavior across environments. Node.js includes the same URL class that browsers do, and you can use it to construct a fully-qualified address.

In the question details you state that window.location.origin is set to a valid URL origin string, so you can use it to resolve URLs. Below is an example that will work in Node.js and in a browser — you can run it in Node.js, or you can copy + paste the code into your dev tools JS console on this page and run it to see the same output. (I'm not posting it as a code snippet because snippets on this site run at a different origin.)

./example.mjs:

/// <reference types="node" />

// According to your question, this is set elsewhere in your environment:
if (!globalThis.window) {
  globalThis.window = {
    location: {
      origin: "https://stackoverflow.com",
    },
  };
}

function resolveUrl(url) {
  return new URL(url, window.location.origin);
}

// A path relative to the current origin:
const url1 = resolveUrl(
  "/questions/75794309/how-does-fetch-detect-if-a-request-using-a-relative-path-is-coming-from-a-browse",
);

// Address at a separate origin:
const url2 = resolveUrl("https://developer.mozilla.org/en-US/docs/Web/API/URL");

console.log("URL 1:", url1.href); // https://stackoverflow.com/questions/75794309/how-does-fetch-detect-if-a-request-using-a-relative-path-is-coming-from-a-browse
console.log("URL 2:", url2.href); // https://developer.mozilla.org/en-US/docs/Web/API/URL

const response = await fetch(url1);

console.log({
  contentType: response.headers.get("Content-Type"), // text/html; charset=utf-8
  docType: (await response.text()).slice(0, 15), // <!DOCTYPE html>
});

In Node.js:

% node --version      
v18.15.0

% node example.mjs
URL 1: https://stackoverflow.com/questions/75794309/how-does-fetch-detect-if-a-request-using-a-relative-path-is-coming-from-a-browse
URL 2: https://developer.mozilla.org/en-US/docs/Web/API/URL
{ contentType: 'text/html; charset=utf-8', docType: '<!DOCTYPE html>' }