Improvements to field set APIs

[sffc]

          We should bikeshed the name of this.

Originally posted by @sffc in #6018 (comment)

[robertbastian]

How about this

Format	UTS-35 symbol	ICU fieldset
Specific non-location	z..zzzz	Z, Zs
Generic non-location	v..vvvv	V, Vs
Offset	O..OOOO	O 1
Location	VVVV	L
Exemplar city	VVV	C
IANA ID	VV	not implemented
BCP-47 ID	V	?
ISO	X..XXXXX	X
aliases	Z..ZZZZZ	-

We got lucky that Z is used for aliases only, so we can use it instead of z. However, using V instead of v means we need to find different fieldsets for all the V symbols. L is the standalone month in UTS-35, and lowercase l is deprecated, which I guess frees it up, C is an "input skeleton symbol" and c is a standalone day-of-week, which I don't think we need either.

Now BCP-47 ID and IANA ID remain. Those will need two different field sets, because the IANA ID will need more data from somewhere (either loading an IanaMapper, or requiring a TimeZoneInfo<WithIana>).

The BCP-47 ID and ISO formats actually have the same bounds, they don't need data and work with a Base timezone info, so we could put them in one field set and discriminate by length, but I'd rather not do that.

Footnotes

1. There's no data diff between O and Os, so I don't think we need Os. ↩

[sffc]

We don't necessarily need to add every style to the field sets, at least not right away. We should prioritize the ones that make sense when combined with datetime. Users who want full access to time zone names can use DateTimePattern and bypass the field sets.

So, for example, the city style probably doesn't need a field set, because "19:00 Paris" doesn't make sense (should be at least "19:00 Paris Time"), and I also left out the ISO format because the Localized Offset format is just better.

[robertbastian]

The fieldsets are not just for combining with a time, they can be used standalone. It's a much nicer API than going through DateTimePattern:

icu4x/components/datetime/examples/timezone_picker.rs

Lines 19 to 20 in 3cad93c

	let non_location_formatter = DateTimeFormatter::try_new(prefs, fieldsets::V::new()).unwrap();
	let city_formatter = DateTimeFormatter::try_new(prefs, fieldsets::X::new()).unwrap();

Also, "19:00 Paris" makes sense, that's why the VVV pattern exists. The ISO format also makes sense in certain situations. I don't understand why we'd hide this functionality.

[sffc]

Field sets are supposed to make it easy to do the right thing. Using VVV instead of VVVV is usually the wrong thing.

If you want a Google Calendar style time zone picker, DateTimePattern is not a bad option, I think.

But if you can design a semantic datetime option that does VVV only when it is "right", and otherwise uses VVVV, I'm all for it.

[sffc]

As suggested in #6029 (comment):

I would support replacing fieldsets::V and friends with fieldsets::zone::GenericNonLocationLong.

Thoughts? @robertbastian @Manishearth

[Manishearth]

I'm in support of clearer names. I don't really have opinions on what those names should be , but it seems like y'all have some clear ideas.

[robertbastian]

Proposal:

// Things we currently have, but exported with new names:
icu::datetime::fieldsets::zone::ExemplarCity
icu::datetime::fieldsets::zone::GenericShort
icu::datetime::fieldsets::zone::GenericLong
icu::datetime::fieldsets::zone::SpecificShort
icu::datetime::fieldsets::zone::SpecificLong
icu::datetime::fieldsets::zone::LocalizedOffsetShort
icu::datetime::fieldsets::zone::LocalizedOffsetLong
(icu::datetime::fieldsets::zone::LocalizedOffsetFixed)
icu::datetime::fieldsets::zone::Location

// We might add other styles when people ask for them.
// These are less human readable, so the use cases as field sets are not obvious
icu::datetime::fieldsets::zone::IanaId
icu::datetime::fieldsets::zone::Bcp47Id
icu::datetime::fieldsets::zone::Iso


// Conversion functions
// currently: YMDT::with_zone_generic(self) -> Combo<YMDT, GenericShort>
// Note: the bound is not _necessary_ but it is useful for documentation purposes
YMDT::with_zone<Z: ZoneMarkers>(self, Z) -> Combo<YMDT, Z>

fieldsets::YMDT::short().zone(fieldsets::zone::SpecificLong)

// Field set expressions, removing the "with":
fieldsets::T::short().hm()
fieldsets::T::short().hms()
fieldsets::T::short().precision(TimePrecision::OptionalMinute)

fieldsets::YMD::short().time(TimePrecision::OptionalMinute).zone(fieldsets::zone::SpecificLong)
fieldsets::YMD::short().hm().zone(fieldsets::zone::SpecificLong)
fieldsets::YMD::short().hms().zone(fieldsets::zone::SpecificLong)

// Consider replacing YMDT with `Combo<YMD, T>`. Also implies `Combo<Combo<YMD, T>, SpecificLong>`

// Scratch pad of other concepts:
fieldsets::YMD::short().time(fieldsets::T::short().hm().with_alignment(...)).zone(fieldsets::zone::SpecificLong)
fieldsets::YMD::with(fieldsets::T.hm()).long().alignment(Alignment::Standard).with(fieldsets::zone::SpecificLong)
fielsets::YMD.long().with(fielsets::zone::SpecificLong)