[DISCUSSION] Reduce amount of ExprCoreTypes

[Yury-Fridlyand]

Is your feature request related to a problem?

1. Do we need so much numeric ExprCoreTypes?
2. -//- datetime ExprCoreTypes?

What solution would you like?
Why not to leave only LONG and DOUBLE or LONG/INT and FLOAT/DOUBLE?
I propose to get rid of BYTE and SHORT types, map byte and short opensearch types to INT. Thus we can reduce amount of math/comparison functions registered and simplify widening rule processing. Further removing of INT and FLOAT and using only LONG and DOUBLE would increase memory usage (for storing those values only!), but increase performance.

BYTE of even BIT/BOOL uses same amount of memory in Java (SQL/JDBC) or even in C (ODBC)(unless you use pragma pack which reduces performance). Refs: one, two.

$ java -Djdk.module.illegalAccess=deny -jar jol-cli.jar internals java.lang.Byte | grep 'Instance size'
Instance size: 16 bytes
$ java -Djdk.module.illegalAccess=deny -jar jol-cli.jar internals java.lang.Integer | grep 'Instance size'
Instance size: 16 bytes
$ java -Djdk.module.illegalAccess=deny -jar jol-cli.jar internals java.lang.Long | grep 'Instance size'
Instance size: 24 byte

Also having both DATETIME and TIMESTAMP ExprCoreTypes is a bit confusing - they contain the same information. Both types don't store time zone information.

What alternatives have you considered?
N/A

Do you have any additional context?
N/A

[acarbonetto]

Also having both DATETIME and TIMESTAMP ExprCoreTypes is a bit confusing - they contain the same information. Both types don't store time zone information.

If #126 is fixed and supports timezone, we could map to a timestamp type with timezone included. Which means we would have a reason to have both a timestamp and datatime type.

Consider not removing datetime.

[Yury-Fridlyand]

#126 is to support more date formats
This ticket to remove duplicated types.
DATETIME and TIMESTAMP contain completely the same info and adding extra complexity

[acarbonetto]

We can remove DATETIME, but there's no real benefit to removing BYTE/SHORT.