precision is not considered when cast value to decimal

[liukun4515]

Describe the bug

If the CastOptions {safe: true} and cast INT32(120) to decimal(3,1), the result should be None, but
we get the result of decimal(Some(1200),3,1)

To Reproduce

Expected behavior

Additional context

[liukun4515]

Do you have any thoughts about this? cc @tustvold @viirya

[tustvold]

I thought we had agreed to only care about overflow of the underlying type, and leave it to users to validate decimals if they care about precision. Provided we aren't overflowing and losing data, I'm not sure why we would validate precision at all?

[liukun4515]

I thought we had agreed to only care about overflow of the underlying type, and leave it to users to validate decimals if
they care about precision. Provided we aren't overflowing and losing data, I'm not sure why we would validate precision at all?

Now arrow-rs has the API of pub fn validate_decimal_precision(&self, precision: u8) -> Result<(), ArrowError> to check the precision for the decimal array.
I think we need a method to do the above requirements, if the value is overflow the precision, and it should be converted to None.
Maybe we should provide the api to change the null_bit_map in the DecimalArray, If user want to validate precision to get None or Error.

@tustvold

[tustvold]

Adding a null_if_overflowed_precision method makes sense to me

[viirya]

Currently we did as @tustvold mentioned. We validate decimals by validate_decimal_precision after casting and put nulls for overflowing slots in an array.

I think it is trivial to do by users. Okay for me if you want to add something like null_if_overflowed_precision to do that.

[liukun4515]

@viirya @tustvold thanks for your advice.
In the user case, some cases want to get the error when the data is overflow for the precision, and some cases don't want to get the error instead of the None for the overflowed slot in the array.