How about ordinals?

[carystanley]

It would be great if intl-messageformat supported ordinal numbers:

1st, 2nd, 3rd, 4th, 10th, 11th, 12th, 13th, 52nd, 53rd, etc...

Here is a really simple version of the logic:

    function ordinal_suffix(num) {
        var s = ["th", "st", "nd", "rd"],
            v = num % 100;
        return num + (s[(v - 20) % 10] || s[v] || s[0]);
    }

I am unsure if ordinal works differently in other languages besides english

[caridy]

@carystanley that's exactly right. To implement ordinals we need to use CLDR to produce the logic per locale. this is not the first time we get this feature requests, we will evaluate it, but it doesn't belong to this repo for sure since ICU message format does not support ordinals, just like it doesn't support relative time, we can also push for that in that front.

[ericf]

I did some research and the CLDR has Rule-Based Number Format (RBNF) data, which includes spellout ("three") and ordinal ("3rd"). As of CLDR v26, these rules are defined in ICU Message syntax — which is just like how our Intl RelativeFormat's data is defined and works.

I see a few ways we could implement ordinals:

1) Add support for selectordinal to IntlMessageFormat:

This would be the easiest to implement, and not add any extra locale data to the library, but it would force the translators to add the correct ordinal suffix inside the optional parts of the messages, e.g.:

Go to the {floor, selectordinal,
  =0 {ground} 
  one {#st} 
  two {#nd}
  few {#rd}
  other {#th}
} floor.

2) Create an IntlRuleBasedNumberFormat lib, use externally:

To support automatic ordinal formatting leveraging the rules built into CLDR, we could create a new IntlRuleBasedNumberFormat, and like IntlRelativeFormat it would be used externally from the ICU Message string; i.e., you would format the value using IntlRuleBasedNumberFormat before passing it to the IntlMessageFormat instance, e.g.:

Go to the {floor} floor.

var msg  = new IntlMessageFormat('Go to the {floor} floor.', 'en');
var rbnf = new IntlRuleBasedNumberFormat('en', {style: 'ordinal'});

var output = msg.format({
    floor: rbnf.format(3)
});

console.log(output); // => "Go to the 3rd floor."

The benefit of this approach is that it doesn't add k-weight to IntlMessageFormat — just like how relative time formatting is separate. The downside is that it's not built into IntlMessageFormat, therefore deviates from the standard ICU Message. That said, at the end of the day, it would still be built into react-intl, handlebars-intl, and dust-intl :-/

Using this external approach in templates would look like this; e.g., in Handlebars:

{{formatMessage (intlGet "messages.floor")
  floor=(formatOrdinal 3)
}}

3) Create an IntlRuleBasedNumberFormat lib, use internally:

The last option would be like option 2) and create a new IntlRuleBasedNumberFormat lib, but also have IntlRelativeFormat depend on it and use it internally — just like it does for Intl.NumberFormat and Intl.DateTimeFormat.

The difference with this approach is that it would add k-weight to the intl-messageformat package both in terms of the runtime, and the locale data. This approach would allow for the ordinal argument type to be used within the ICU Message strings; e.g.:

Go to the {floor, ordinal} floor.

Edit: The ICU Message syntax defines ordinal as an argument type; we just don't currently support it in our implementation.

---

What do you guys think?

[juandopazo]

It looks like Intl.MessageFormat needs a plugin system.

[caridy]

@juandopazo we don't need a plugin system, we just need to convince ICU folks that {floor, ordinal} makes sense, just like {floor, number} does it :)

@ericf my vote goes to option 2, which will pave the way to option 3, just like we did for relative, and we can engage with ICU to try the waters there.

[ericf]

@caridy @juandopazo I wasn't clear before, but ordinal is an argument type defined by the ICU Message syntax; we just don't currently support it in our implementation. I edited option 3 above to make this clear. So there's no need to convince the ICU folks, they already support it 😄

If you look at the details of the ICU Message syntax, you'll see they have support for selectordinal, ordinal, and spellout argument types — all of which we don't currently support. Note that choiceStyle and ChoiceFormat have been deprecated.

[caridy]

awesome, lets plan for that then, we need an implementer now :)

/cc @lwelti

[ericf]

So we can stage this by first easily adding support for selectordinal arguments, option 1 above. @carystanley do you think that would work to unblock you?

Then we work on the automatic ordinal argument formatting via a new IntlRuleBasedNumberFormat lib, which will take a lot more time and effort for us to implement.

[srl295]

By the way, concerning ICU please reach out to us if needed, not so hard to convince, and code does talk :) icu-project.org

[caridy]

well, they can do that today without anything new:

Go to the {floor, select,
  =0 {ground} 
  one {#st} 
  two {#nd}
  few {#rd}
  other {#th}
} floor.

not ideal, but at least this unblock them.

[caridy]

@srl295, lets start the conversation about relative time, e.g.: {dateValue, relative}

[srl295]

@caridy relative time sounds good to me, but probably best started at http://bugs.icu-project.org/trac/newticket . Apologies if you knew this, but the select format in ICU was itself a Yahoo! contribution. Actually there's this ticket (note: follow the trail of 'xref' fields to find related tix) that might be interesting.

edit on consideration, I think a new ticket would be best, and reference any relevant ones you dig up, or even this thread. This is a great time to suggest changes in msgformat.

[caridy]

ICU ticket for relative time in messages: http://bugs.icu-project.org/trac/ticket/11409

[srl295]

@caridy 👍 . added a backlink here and an xref

[srl295]

@caridy and, the ICU ticket was moved to unscheduled. Is someone interested in contributing a C++ and Java design and implementation?