(Adding comment again, so we can discuss here)

@rsolomakhin, I've tried to summarize what you said about how Chrome does the ordering, while keeping the order in which things are matched optional. However, I added matching on the document's body's language as first, to match what we say in .show(). The "Any other criteria the user agent deems suitable" if for the fallback you described, whereby you pick the first available.

WDYT? Would this work?

This works.

@mnoorenberghe, wdty?

I'm not sure what the specific question is… do you want me to comment on the locale ordering or the spec change in general? For the locale ordering, as long as the UA isn't required to show the PR dialog in a language that the UA doesn't have installed then I'm fine with it.

On a separate note, I wonder whether we really need to derive region from the regionCode… it would be simpler to also return the ISO3166-2 code as the region.

One concern I have with the spec change is that I've never seen ISO3166-2 used with the country code prefix in the wild. As a web developer I would expect regionCode to be "CA" for California, not "US-CA" if I hadn't read documentation on the API and only saw the webidl FWIW. I'm not sure if there are exceptions to cutting off the first 3 characters in order to get the region code specific to the country. If there is then this will likely cause issues for merchants when mapping the PR's address to existing systems.

As a web developer I would expect regionCode to be "CA" for California, not "US-CA"

With the information I have at the moment, I'm leaning towards just returning the ISO3166-2 format. It might save us from having to deal with special cases where there is ambiguity, but I also don't know if there are any. I'll try to do a bit more reading.

I'm not sure if there are exceptions to cutting off the first 3 characters in order to get the region code specific to the country.

I don't think there are, from my reading of the ISO3166-2 spec. But I'll take a closer look as it's been a few weeks.

Coming back to this and re-reading the ISO spec, I can't determine that there would be significant issues with chopping off country and dash. I'm still reluctant to do it tho. I'd rather just stick to the full code "just in case" and just so we are not creating new conventions. @mnoorenberghe, is that ok?

Although I'm always hesitant to parse other people's strings, ISO 3166-2 is defined to use ISO 3166-1 alpha-2 code elements as the first two characters of the string, followed by the hyphen character, followed by the region itself. So I think we'd be safe in truncated the first three chars.

I guess it does also remove a source of ambiguity/redundancy, as .country will now need to be there to derive .region. Do we still attempt to do validation on a synthetic 3166-2 code? I would guess no: we attempt to populate region iff county and regionCode are both there and they both together form a known 3366-2 code. Can work with that.

@domenic, stop reading this! get back to vacation! 🍹

But when you are back... by design, address.region is overridable by the developer. So, eventually:

const dict = {
   country: "PT",
   regionCode: "PT-11",
}
const address = new PaymentAddress(dict);
address.region; // "Lisboa", per ISO ISO3166-2

// And... then.. 
const dict = {
   country: "PT",
   regionCode: "PT-11",
   region: "Lisbon",
}
const address = new PaymentAddress(dict);
address.region; // Lisbon, per developer override.

Hmm.... now I'm wondering if we should allow some members to be nullable on the AddressInit dictionary:

const dict = {
   country: "PT",
   regionCode: "PT-11",
   region: null, // or undefined, means "derive it for me" 
}

While:

const dict = {
   country: "PT",
   regionCode: "PT-11",
   region: "", // trash it ... I'll handle it
}

@domenic, if you are back from vacation, #690 (comment)

I think undefined ("not present" in dictionary-speak) is the right signifier. So basically only execute this step if details["region"] does not exist. Otherwise set address.[[region]] to details["region"].

Whether we want to allow details["region"] to be nullable or not, as some kind of explicit "don't derive this for me, but also make it not a real region value"---I'm not sure.

Should we consider deriving region from regionCode, so the caller only has to pass one? I'm thinking ahead to the constructor; do we want to prevent new PaymentAddress({ ..., regionCode: "PT-11", region: "Tasmania" })?

I guess for the cases where there is no corresponding regionCode, we'd still want to pass region. But if regionCode is not the empty string, deriving region from it might be good...

But if regionCode is not the empty string, deriving region from it might be good...

Agree. It's definitely doable. ISO3166-2 provides lookup tables, for example:

AD-07* Andorra la Vella
AD-02* Canillo
AD-03* Encamp
AD-08* Escaldes-Engordany
AD-04* La Massana
AD-05* Ordino
AD-06* Sant Julià de Lòria

Where column 2 (defined thing) is "...the country subdivision names in the administrative language of the country concerned, where relevant with diacritic signs..." (as Unicode). Note that "a country’s administrative language is a written language used by the administration of the country at the national level". So, the result won't be in English a lot of the time - but that's fine, IMO. This is not for display purposes.

So, something like:

The steps to derive a region from a country subdivision code element are as follows. The string takes a DOMString subCode as input and returns a DOMString.

1. If validate a region code subCode return false, throw RangeError exception.
2. ASCII uppercase subCode, and let normalizeCode be the result.
3. Find normalizeCode in Section 8, "List of country subdivision names and their code elements" of ISO3166-2 and return the value of "column 2" for the matching country subdivision code element.

Sound good?

That sounds great! Probably we should get implementer buy-ins though; maybe as a separate PR? Because it'd mean having to ship that table with the browser...

Agree... ☝️@rsolomakhin, @mnoorenberghe, something to start thinking about. Would appreciate your input.

We are already using libaddressinput (and I believe Chromium does too) so it would be great if we could also use that here instead of having two versions of similar data. Does ISO3166-2 only have one name per region? What about when there are multiple official languages? Consider CA-QC: is it "Quebec" or "Québec" in ISO3166-2. libaddressinput provides both but without a clear way to know which one would match ISO3166-2.

Presumedly, we can just grab the "en" column 2

I'm not quite sure how this all fits together but won't this affect what gets returned in the region property and therefore may be displayed back to the user on the payment confirmation page the merchant renders? If so, it would seem like we should use the language best for the user.

We can probably perform some kind of lookup, based on the document language. But then it gets into preference order for when it doesn't match the user's language (and there is no "en").

Seems crazy town, but:

1. Try to match document language.
2. Try to match user language (like Accept-Language)?
3. Try to match "en".
4. Um... just give me whatever you got.

We use libaddressinput in Chromium and @sebsg is helping us to add the ISO codes to that.

But if regionCode is not the empty string, deriving region from it might be good.

The regionCode field takes care of interop, so we don't necessarily have to agree on how the region field is derived from that. Having said that, the rule of thumb we have been using in Chrome is to use the browser user interface language when deciding between Quebec and Québec. If the browser user interface language does not match any of the languages supported by the country, then Chrome falls back to the first language specified in libaddressinput data for that country. For example, if the browser user interface language is Italian, then Canadian addresses will be in English. Although the default language works out to be English for Canada, that's not something that we should hardcode for all countries in the world, IMHO.