Bikeshed: What should neo datetime placeholder markers be called

[Manishearth]

Part of #3865

From #4185

We have an opportunity to reduce data size for calendars that don't need, say, day symbols data (or numeric month formatting data). For generic and AnyCalendar formatting to work, however, we need to be able to abstract over calendars that both have and don't have this data alike. Ideally we can do something like type DaySymbols = (), however this causes trouble when we attempt to abstract over this with traits since you can't easily do "or" bounds with traits (here we need "Yokeable = LinearSymbolsV1 OR Self = ()").

We should have an always-empty key for this, like PlaceholderDaySymbolsV1, which we never attempt to load but exists to satisfy the trait bound.

We should figure out what to call it.

(In general I'd like to avoid the term "dummy", it's often considered non-inclusive)

[sffc]

Discuss with:

• @echeran
• @younies
• @sffc
• @Manishearth

[Manishearth]

Approach:

• Have PlaceholderDaySymbolsV1 => LinearSymbolsV1 marker type, name TBD
• In DTF data loading skip loading data when you encounter CldrCalendar assoc types that are placeholder. Either use type_id or something else.

Name bikeshed:

• PlaceholderDaySymbolsV1
• EmptyDaySymbolsV1
• EmptyMarker<Yokeable>

Decision: EmptyMarker<Yokeable>

impl<Y: Yokeable> KeyedDataMarker for EmptyMarker<Y> {
    const KEY: DataKey = data_key!("_empty", singleton);
}

// on most providers
impl<Y: Yokeable> DataProvider<EmptyMarker<Y>> for Foo {
    debug_assert!(false);
    return DataError::MissingDataKey.with_key(EmptyMarker<Y>::KEY);
}


// Check for the EmptyMarker before calling .load()
// this should be a convenience method on EmptyMarker
if core::any::TypeId::of::<M>() == core::any::TypeId::of::<EmptyMarker<M::Yokeable>>() {
    // skip loading
}

Discussion:

• rob: should not need a new error variant, since it's not something to catch and handle
• manish: also want same behavior for blob.
• manish: ultimately should never be called (and documented as such)
• sffc: debug assert in AsDeserializingBufferProvider?
• manish: yes

Agreed: @Manishearth @robertbastian @sffc

[sffc]

I need this for skeleta, so I am taking it now

[sffc]

Since the key fails to load, instead of EmptyMarker, wdyt of NeverMarker?

[Manishearth]

works!