Possibly-broken ToASCII test cases

[domenic]

I'm working on updating jsdom's TR46 implementation for the Unicode 16.0.0 / TR46 rev 33 changes, and I believe I've found some incorrect test cases.

toascii.json (all from WebKit upstreaming in 6fa3fe8)

@annevk

  {
    "input": "www.lookout.net\u2A7480",
    "output": null
  },

My implementation converts this to www.lookout.net::=80.

  {
    "input": "www\u00A0.lookout.net",
    "output": null
  }

My implementation passes this through as valid. I cannot find anything in https://unicode.org/reports/tr46/#Validity_Criteria which would disallow it.

  {
    "input": "\u001flookout.net",
    "output": null
  }

My implementation passes this through as valid. This would be invalid if we were passing UseSTD3ASCIIRules=true, but I don't believe we are. At least, my test harness never has, and the URL Standard passes it as false (via setting beStrict to false).

IdnaTestV2.json (from d0cd7c0)

@rmisev

  {
    "comment": "V4; V2 (ignored)",
    "input": "xn--xn--a--gua.pt",
    "output": null
  },

My implementation gives back the input for this. I haven't yet dug into where the bug might be here. It's possible I didn't adapt my code well enough.

  {
    "comment": "V4; V2 (ignored)",
    "input": "xn--xn---epa",
    "output": null
  },

My implementation gives back the input for this. I haven't yet dug into where the bug might be here. It's possible I didn't adapt my code well enough.

[annevk]

• www.lookout.net::=80 Why would this not fail on https://url.spec.whatwg.org/#forbidden-host-code-point as checked by https://url.spec.whatwg.org/#concept-domain-to-ascii step 3.2?
• www\u00A0.lookout.net What is the Punycode output then?
• \u001flookout.net This would fail for the same reason as the first example here. C0 is not allowed.

[domenic]

• Why would this not fail on https://url.spec.whatwg.org/#forbidden-host-code-point as checked by https://url.spec.whatwg.org/#concept-domain-to-ascii step 3.2?

Good catch. I didn't implement whatwg/url@c3d173f yet.

Unfortunately this is the first time toascii.json is not usable for testing TR46's ToASCII, but instead of the URL Standard's domain-to-ASCII. That's a pain for my project setup which separates those.

Do you think it'd be reasonable to add a boolean to the test cases saying that they are not applicable to TR46?

• www\u00A0.lookout.net What is the Punycode output then?

It doesn't get Punycoded. It appears it gets converted to U+0020. So I guess this is the same issue.

[annevk]

Sounds reasonable to me.

[domenic]

IdnaTestV2.json (from d0cd7c0)

These cases are my bad. They come from whatwg/url#803. I had this code:

  // "4. If not CheckHyphens, the label must not begin with “xn--”."
  // Disabled while we figure out https://github.com/whatwg/url/issues/803.
  // if (!checkHyphens) {
  //   if (label.startsWith("xn--")) {
  //     return false;
  //   }
  // }

After un-commenting this out, they pass.

I'll close this issue, but quick review on #51371 would be appreciated.