Inconsistent type inference for date add/subtract interval operations

[LeonidChistov]

Following operations are defined in Substrait prototypes:

add(date, interval_year): -> timestamp
add(date, interval_day): -> timestamp
subtract(date, interval_year): -> date
subtract(date, interval_day): -> date

Is there a reason why add and subtract should behave differently wrt to result type?

[EpsilonPrime]

I don't believe there is a reason. The add functions taking a date should return a date (the same type as the input).

[LeonidChistov]

IMHO, following type inference would make more sense:

add(date, interval_year): -> date
add(date, interval_day): -> timestamp
subtract(date, interval_year): -> date
subtract(date, interval_day): -> timestamp

Because otherwise there is a high chance that we just loose "interval_day" value except for the "day" part.

But from the other point of view, someone who does not want to loose precision, may cast date to timestamp before adding/subtracting interval.

Do you think it make sense to create Issue based on that discussion?

[EpsilonPrime]

Your analysis makes sense, it should be timestamp. I did find an old PR (#337) that corrected subtract's type, presumably they missed add at that time.

We could go with an issue or straight into a PR.

[LeonidChistov]

#576