[PROTOCOL RFC] Add `TIME` & `INTERVAL` as supported dtypes

[ion-elgreco]

I am often getting the question, why doesn't delta-rs support time/interval dtypes.. Seems obvious to add since it's supported by Parquet and supported by most engines.

We can do the same thing as timestampNtz and gate these behind a reader & writer feature. I am proposing we add these primitive types:

Type Name	Description
time	Microsecond precision time of day (backed by parquet TimeType) https://github.com/apache/parquet-format/blob/master/LogicalTypes.md#time
interval	A fixed amount of time https://github.com/apache/parquet-format/blob/master/LogicalTypes.md#interval

Time

This feature introduces a new data type to support time. For example: 00:00:00, or 24:00:00. The serialization method would be: {hour}:{minute}:{second}

Interval

This feature introduces a new data type to support day-time intervals. For example: 5d 10h 5m 10s. The serialization method would be: {day}d:{hour}h:{minute}m:{second}s

To support these feature:

To have a column of Time/Interval type in a table, the table must have Reader Version 3 and Writer Version 7. A feature name time/interval must exist in the table's readerFeatures and writerFeatures.

Engine support:

Engine	Type Name	Support	equivalent engine type
Spark	time	No	N/A
Spark	interval	Yes	DayTimeIntervalType
Trino	time	Yes	TIME(P)
Trino	interval	Yes	INTERVAL DAY TO SECOND
Flink	time	Yes	TIME(P)
Flink	interval	Yes	INTERVAL DAY TO SECOND
Datafusion	time	Yes	Time (in arrow: Time64)
Datafusion	interval	Yes	Interval(DayTime)

[ion-elgreco]

Supersedes this issue: https://github.com/delta-io/delta/issues/2319

@tdas @ryan-johnson-databricks and @bart-samwel

[bart-samwel]

This makes sense to me if it's just like TimestampNTZ. It's unfortunate that Spark doesn't have a TIME type (yet). :/

[ion-elgreco]

@bart-samwel what would be the next steps to make this happen?

[scovich]

Seems like we need to add the datatype support to spark, as a starting point? (seems quite doable, but I'm not the expert there)

[ion-elgreco]

Seems like we need to add the datatype support to spark, as a starting point? (seems quite doable, but I'm not the expert there)

Can we maybe start with interval first then, since it's supported in all engines?

I can add the support in delta-kernel-rs and Delta-RS for both types

[ion-elgreco]

@scovich @tdas @ryan-johnson-databricks and @bart-samwel

Can we move forward with this adding it to the protocol?

@scovich I don't see why we should at this to Spark first