[threeten-dev] Clarification request: DateTimeFormatter.ofLocalizedDate/Time behavior

[Masayoshi Okutsu]

On 2/4/2013 7:49 PM, Stephen Colebourne wrote:
> On 4 February 2013 06:05, Masayoshi Okutsu <masayoshi.okutsu at oracle.com> wrote:
>> What's the expected behavior of the following code?
>>
>>    DateTimeFormatter dtf =
>> DateTimeFormatter.ofLocalizedDate(FormatStyle.FULL);
>>    System.out.println(dtf.format(JapaneseChronology.INSTANCE.dateNow()));
>>
>> Currently, the format pattern is taken from a DateFormat instance if it's
>> actually a SimpleDateFormat. The DateFormat.get*Instance behavior is
>> affected by the specified Locale for selecting the calendar system to use.
>> For example, if the default Locale is "ja-JP-u-ca-japanese", the format
>> pattern from a SimpleDateFormat is for the Japanese calendar system. But
>> that doesn't seem to be the intended behavior (relying on the locale's
>> calendar).
> The current implementation is in DateTimeFormatStyleProvider. It has a
> "TODO" to make it better.
>
> The code currently uses the locale of the formatter to perform the
> lookup, and the chronology (which is available) is unused.
>
> Ideally, DateTimeFormatStyleProvider should be reworked entirely to
> access the CLDR data more directly.

That's what I'm trying to do.

> So, given this code:
> str = DateTimeFormatter.ofLocalizedDate(FormatStyle.FULL).withLocale(loc).format(chronoDate);
>
> I would expect that both the chronology of the date and the specified
> locale would influence the pattern that is picked, and thus the text
> output. The chronology should be used to pick the calendar system of
> the pattern, and the locale should be used to pick the English vs
> French vs Japanese view of the pattern (or even a different pattern).
>
> My feeling is that this could be implemented by changing the locale in
> DateTimeFormatStyleProvider to add the "ca" variation from the
> chronology object.

So is the intention to evaluate the Chronology from chronoDate and the 
Locale from a DateTimeFormatter on every format() call? If so, 
DateTimeFormatterbuilder needs some changes. Currently, the Chronology 
is taken from a DateTimeFormatter, which is always ISO.

> JapaneseDate + locale jp_JP => locale jp_JP-u-ca-japanese
>
> However, you raise an interesting question of what if the locale
> includes a "ca" block already. The simple case is:
>   JapaneseDate + locale jp_JP-japanese => locale jp_JP-u-ca-japanese
>
> the complex cases are:
>   JapaneseDate + locale jp_JP-minguo => ?
>   MinguoDate + locale jp_JP-japanese => ?

I think any calendar type given by a Locale should be ignored in 
DateTimeFormat. Otherwise, it'd be too confusing.

Thanks,
Masayoshi

>
> I think you have more knowledge than me to decide on behaviour there
> (taking into account what data is available in CLDR)
>
> Stephen