Support decimal encoding of NaN and Infinities values

[wendigo]

No description provided.

[findepi]

These are valid values for Trino NUMBER but not for DECIMAL.
If we return NaN for decimal, this should fail somewhere later. I think for DECIMALs it's better to keep failing here

[wendigo]

Since decimal is shared between decimal and number I think it's better to fail in Trino once nan or infinity is returned for Decimal

[findepi]

Let's have a test for that:

• Fail correctly when Python UDF decimal result overflows trino#28925

[findepi]

I looked more into this. I see that Python's decimal.Decimal really corresponds well to Trino's NUMBER type. Match better than to Trino's DECIMAL. So conceptually this change makes sense.

However, TrinoType enum is what it is -- enum of all trino types.
Mapping on Trino side a NUMBER to a DECIMAL feels wrong, one would do the other way around (NUMBER is a strictly wider type than DECIMAL).

If we do not want have more entries in TrinoType enum, then I think we should rename the existing one (DECIMAL → NUMBER).

[wendigo]

I can add a separate NUMBER type but it's handling will be similar to the decimal

[findepi]

SGTM

[wendigo]

@findepi which version? ;)

[findepi]

this one:

• Add dedicated NUMBER type #12