NO_PROXY=example.com does not bypass subdomains for http.request(), unlike fetch()

从 repro.js 的行为和 http 入口 ProxyConfig#shouldUseProxy 开始；将其与 fetch() 的 EnvHttpProxyAgent 以及 NO_PROXY 文档进行比较。完成的标准是：一个普通的 example.com 条目会在标签边界处绕过其子域名，并且 http.request() 与 fetch() 的行为一致。

Version

v24.18.1, also on main (v27.0.0-pre)

Platform

Darwin xxx.local 25.3.0 Darwin Kernel Version 25.3.0: Wed Jan 28 20:54:46 PST 2026; root:xnu-12377.91.3~2/RELEASE_ARM64_T6000 arm64

Subsystem

http

What steps will reproduce the bug?

Dead proxy on port 1 — the error code shows where the request went:

NODE_USE_ENV_PROXY=1 HTTP_PROXY=http://127.0.0.1:1 NO_PROXY=example.com node repro.js

const http = require('node:http');

http.get('http://internal.example.com/').on('error', (e) => console.log(e.cause?.code ?? e.code));
// ECONNREFUSED — went through the proxy

fetch('http://internal.example.com/').catch((e) => console.log(e.cause?.code ?? e.code));
// ENOTFOUND — bypassed, hit DNS

How often does it reproduce? Is there a required condition?

Always, when the NO_PROXY entry has no leading dot and the request host is a subdomain of that entry.

What is the expected behavior? Why is that the expected behavior?

Same NO_PROXY in the same process should mean the same thing for http.request() / http.get() and fetch(). http.setGlobalProxyFromEnv() and the docs present one NO_PROXY format for both.

fetch() (undici EnvHttpProxyAgent) already treats a plain example.com as the host and its subdomains, with a label boundary so notexample.com does not match. http should do the same.

What do you see instead?

ProxyConfig#shouldUseProxy exact-matches a plain entry against the hostname. NO_PROXY=example.com therefore does not bypass internal.example.com for http.request(), while fetch() does.

The docs currently describe this as “Exact host name match”, which matches the http implementation and not fetch().

Additional information

Refs: #57872, #62907