src: do not ignore IDNA conversion error

[TimothyGu]

Currently, the ICU-based IDNA conversion methods only return errors on those passed along through a UErrorCode. However, according to ICU's documentation for uidna_nameToASCII(),

If any processing step fails, then pInfo->errors will be non-zero and the result might not be an ASCII string. The domain name might be modified according to the types of errors. Labels with severe errors will be left in (or turned into) their Unicode form.

The UErrorCode indicates an error only in exceptional cases, such as a U_MEMORY_ALLOCATION_ERROR.

In other words, when non-catastrophically invalid domains are passed, ToASCII() and ToUnicode() (and their downstream url.domainToASCII() and url.domainToUnicode()) currently return garbled domain names instead of errors.

This PR makes the C++ binding methods report errors in pInfo->errors in addition to UErrorCode, thereby fixing those aforementioned problems.

Also included in this PR are additional tests for invalid situations as well as documentation clarifications for the user-facing url.domainToASCII() and url.domainToUnicode().

Before vs. after

> url.domainToASCII('\ufffd.com')
'�.com'
> url.domainToUnicode('xn---\x03.com')
'xn---\u0003.com'
> process.binding('icu').toASCII('\ufffd.com')
'�.com'
> process.binding('icu').toUnicode('xn---\x03.com')
'xn---\u0003.com'
> process.binding('icu').toUnicode('xn--- .com')
'xn--- .com'

> url.domainToASCII('\ufffd.com')
''
> url.domainToUnicode('xn---\x03.com')
''
> process.binding('icu').toASCII('\ufffd.com')
Error: Cannot convert name to ASCII
    at repl:1:24
> process.binding('icu').toUnicode('xn---\x03.com')
Error: Cannot convert name to Unicode
    at repl:1:24
> process.binding('icu').toUnicode('xn--- .com')
Error: Cannot convert name to Unicode
    at repl:1:24

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

Affected core subsystem(s)

[TimothyGu]

CI: https://ci.nodejs.org/job/node-test-pull-request/6585/

[TimothyGu]

Hopefully the issue with legacy url parser is fixed.

/cc @nodejs/intl @nodejs/url

New CI: https://ci.nodejs.org/job/node-test-pull-request/6586/

[addaleax]

Btw, should this be deserialization, and mention that it is the inverse of domainToASCII?

[TimothyGu]

It is serialization, since the domain is fully parsed and subsequently serialized from the parsed form. It's just that it uses a different algorithm for deserialization.

[addaleax]

Can you update the args.Length() check above to use 2? Also, you probably want to add a CHECK(args[1]->IsBoolean()); or do args[1]->BooleanValue() instead.

[TimothyGu]

I didn't update the check for argument length, since (as the comment is trying to say) it is an optional argument, so that existing usage of toUnicode(str) would still work. V8 automatically returns an Undefined for out-of-range args[] dereference.

Wasn't aware of BooleanValue(). Will use that instead.

[addaleax]

(ditto)

[addaleax]

Is this error part of any non-experimental API? Could we change it to Cannot encode name to ASCII as Punycode?

[TimothyGu]

Yes for toASCII

> url.parse(`http://${'é'.repeat(230)}.com/`)
Error: Cannot convert name to ASCII

[addaleax]

Is this error part of any non-experimental API? Could we change it to Cannot decode name as Punycode? (basically the same question I also posted below).

[TimothyGu]

No; in fact the toUnicode JS function isn't used in the code base at all. Maybe we should just remove this method?

/cc @jasnell

[bnoordhuis]

If it's not used, it can be removed.

[jasnell]

Remove which function specifically? The `i18n::ToUnicode' function is definitely used.

[TimothyGu]

@jasnell, the exposed process.binding('icu').toUnicode() JS function.

[addaleax]

Does this compile? Seems like the env->context() argument is missing

[TimothyGu]

@addaleax, you are right. Forgot to push fde77b3

[bnoordhuis]

If it's not used, it can be removed.

[joyeecheung]

This should also fix the missing errors when parsing percent-encoded disallowed characters in hosts(https://github.com/nodejs/node/blob/master/test/fixtures/url-tests.js#L4499) since we are no longer ignoring UIDNA_ERROR_DISALLOWED, you can turn them on in this PR if you like.

[jasnell]

CI: https://ci.nodejs.org/job/node-test-pull-request/6605/

[TimothyGu]

@jasnell, did you see #11549 (comment)?

[TimothyGu]

Test re-enabled per @joyeecheung. Will land tomorrow.

CI: https://ci.nodejs.org/job/node-test-pull-request/6619/

[TimothyGu]

Landed in a520508...7ceea2a. toUnicode() isn't removed for now, since it provides equivalence to the punycode module, and though unused in the code base it is well-tested.