Node: DNS resolution fails for internationalized domain names

[saghul]

• Version: v10.14.2
• Platform: Darwin november.local 18.2.0 Darwin Kernel Version 18.2.0: Mon Nov 12 20:24:46 PST 2018; root:xnu-4903.231.4~2/RELEASE_X86_64 x86_64
• Subsystem: dns, cares_wrap

Currently the DNS module seems to utf-8 encode the names passed to c-ares for DNS resolution: https://github.com/nodejs/node/blob/2c0a75118cd6a2eaf6b45fe8c67336f44c86ab0f/src/cares_wrap.cc#L1803

That doesn't work for internationalized domain names, which need to be IDNA encoded.

DNS is pretty much all ASCII, so IMHO Node shouldn't utf-8 encode on input.

Here is a list of international domain names to test.

This will produce an error:

> dns.resolve('españa.icom.museum', function(r, e) { console.log(r, e) })`
> { Error: queryA ENOTFOUND españa.icom.museum
    at QueryReqWrap.onresolve [as oncomplete] (dns.js:197:19)
  errno: 'ENOTFOUND',
  code: 'ENOTFOUND',
  syscall: 'queryA',
  hostname: 'españa.icom.museum' } undefined

Whereas the proper IDNA encoded version works:

> dns.resolve('xn--espaa-rta.icom.museum', function(e, r) { console.log(e, r) })
> null [ '91.194.60.138' ]

[targos]

All supported versions of Node are affected. I'm looking into it.

[santigimeno]

Possible fix in https://github.com/nodejs/node/pull/25559

[silverwind]

I'm quite surprised that this affects so many version. I'm pretty sure IDN did work at least at some point in time.

[bnoordhuis]

@santigimeno and I opened PRs but now that I think about it, this issue should be fixed once v10.x upgrades to libuv v1.24.0 because that's the first libuv release that includes https://github.com/libuv/libuv/pull/2046.

Landing the PRs would still be useful for:

1. v6.x (and v8.x?), because libuv isn't upgraded on that branch, and
2. z/os, because EBCDIC

[bnoordhuis]

I'm pretty sure IDN did work at least at some point in time.

@silverwind If you'll allow me to self-quote from https://github.com/nodejs/node/pull/25679#issue-247231927:

[..] Node.js left it up to the system resolver or c-ares.

Leaving it to the system resolver introduces platform differences
because:

    some support IDNA 2008
    some only IDNA 2003 (glibc until 2.28), and
    some don't support IDNA at all (musl libc)

In other words, Node's behavior is currently platform-dependent (bad.)