feat: Add timestamp arithmetic functionality

[jzaralim]

Description

This PR adds the functions timestampadd and timestampsub. Example usages are:

TIMESTAMPADD(MINUTES, 40, time)
TIMESTAMPSUB(HOURS, 10, time)

In order to do this, a new "type" was introduced, INTERVAL, that represents an interval of time. Users define them using an IntervalExpression, which consists of numeric expression and a WindowUnit. INTERVAL can only be used within a function and cannot be used as an actual data type (similar to intervals in MySQL).

Testing done

Unit/QTT tests for each of the functions

Reviewer checklist

• Ensure docs are updated if necessary. (eg. if a user visible feature is being added or changed).
• Ensure relevant issues are linked (description should include text like "Fixes #")

[JimGalasyn]

LGTM, with a few suggestions.

[blueedgenick]

Hey @jzaralim - apologies if this was already debated, perhaps in the klip, and I simply missed it - I'm thinking that it might be easier for our users, and more in line with other common systems, to have a datediff function rather than specific add/subtract-to-timestamp functions ? In combination with the neat new interval building block you have here it can be composed into many interesting permutations. It's pretty common to use this kind of function (and supply a negative interval value to go back in time, positive to go forwards) and of course the naming suggests it may one day work for any kind of date/time data (we can be flexible or not on that as the capabilities get richer).

[spena]

@blueedgenick Doesn't datediff give a different result than timestampadd for instance? The idea of the timestampadd is to add units to a timestamp and return a timestamp. But datediff returns an integer instead. That seems different, doesn't it?

[jzaralim]

@blueedgenick If you're thinking of a function that generically adds/subtracts an interval of time from a timestamp, then currently there isn't anything stopping an interval from having a negative value, so timestampadd(date, -1 day) would actually return the same thing as timestampsub(date, 1 day). I haven't seen any MySQL/Snowflake etc function names that imply this though.

[spena]

Cool, thanks @jzaralim. I left some comments. Btw, I'm not sure if using interval as another type. Is there another way to avoid that? Can we specify the right syntax in SqlBase.g4 instead?

Btw, I looked at other DBs, and they use the INTERVAL keyword in their interval parameters. For instance, date_add(time, INTERVAL '5 hours 5 minutes'). Snowflake, Postgres and Mysql support it that way. Do you think we should do it too?

[spena]

Is there a way to change this error message? no viable alternative at ... does not look helpful for users. If the problem is generic for other functions, then we can ignore it for now.

[spena]

Any way to change the error message? Users may get confused about which 5 parameter is.

[spena]

Is this INTERVAL string printed somewhere? I wonder if you were adding an INTERVAL keyword, but I didn't not see it.

[jzaralim]

Cool, thanks @jzaralim. I left some comments. Btw, I'm not sure if using interval as another type. Is there another way to avoid that? Can we specify the right syntax in SqlBase.g4 instead?

Btw, I looked at other DBs, and they use the INTERVAL keyword in their interval parameters. For instance, date_add(time, INTERVAL '5 hours 5 minutes'). Snowflake, Postgres and Mysql support it that way. Do you think we should do it too?

The main reason interval is its own type is because we need the UDFs to be able to differentiate between regular integers and intervals. SqlLambda exists now though, so it makes sense to refactor intervals to follow the same pattern.

I'm not too sure what purpose the INTERVAL keyword would have, but since it is a well known syntax, let's do that too. It should be a simple change anyway.

[jzaralim]

@spena I've changed the grammar to use a separate time unit instead of the interval type from before. I will be rebasing this after #7103 merges though, since it touches a lot of the same code.

[spena]

LGTM. Thanks @jzaralim