Add Intl.Locale test for language getter where language subtag is "und"

[lionel-rowe]

Currently, both Chromium (bug) and Safari (bug) fail the following:

assert(new Intl.Locale('und').language === 'und')
assert(new Intl.Locale('und-US-u-co-emoji').language === 'und')

Firefox has the correct behavior.

Could be useful to have a test for this specific case.

[ptomato]

I'm not sure if Firefox's is the correct behaviour. I'm looking at Canonical Unicode Locale Identifiers and I don't understand all of the terms there, but I do note that it has a special case to exclude und from one of the steps. Maybe someone with more expertise (like @gibson042 or @sffc) would be able to say what the correct behaviour is.

However, it does seem to me that Chrome and Safari's behaviour is incorrect. The loc.[[Locale]] slot being undefined fails several assertions in the spec, so it should not be possible to return undefined from this property getter.

[anba]

but I do note that it has a special case to exclude und from one of the steps.

That's probably happening for "Unicode CLDR locale identifier", but ECMA-402 uses "Unicode BCP 47 locale identifiers".

[gibson042]

I'm not sure if Firefox's is the correct behaviour. I'm looking at Canonical Unicode Locale Identifiers and I don't understand all of the terms there, but I do note that it has a special case to exclude und from one of the steps. Maybe someone with more expertise (like @gibson042 or @sffc) would be able to say what the correct behaviour is.

However, it does seem to me that Chrome and Safari's behaviour is incorrect. The loc.[[Locale]] slot being undefined fails several assertions in the spec, so it should not be possible to return undefined from this property getter.

The [[Locale]] slot of a Locale instance must always be a Unicode canonicalized locale identifier string, which for input "und" is "und" and for input "und-US-u-co-emoji" is "und-US-u-co-emoji" (i.e., both are already in canonical form). The baseName, language, script, and region getters extract and return relevant substrings. In short, the asserts in this report must pass in any conforming implementation of ECMA-402.

[ptomato]

OK, thanks!