UnscaledValue¶

Overview¶

The UnscaledValue expression extracts the unscaled Long value from a Decimal data type. This is an internal expression created only by the Catalyst optimizer to efficiently access the underlying long representation of decimal values without performing scaling operations.

Syntax¶

This is an internal expression not directly exposed in SQL or DataFrame API. It is generated automatically by the Catalyst optimizer during query planning and optimization phases.

Arguments¶

Argument	Type	Description
child	Expression	The input expression that must evaluate to a Decimal type

Return Type¶

LongType - returns the unscaled long value of the input decimal.

Supported Data Types¶

DecimalType (all precision and scale combinations)

Algorithm¶

Takes a Decimal input value and extracts its internal unscaled Long representation
No type checking is performed since this is an optimizer-generated expression
Directly calls the toUnscaledLong() method on the Decimal object
Preserves null values through null-safe evaluation
Uses efficient code generation for runtime execution

Partitioning Behavior¶

This expression preserves partitioning characteristics:

Does not require shuffle operations
Maintains the same partitioning as the input expression
Can be safely pushed down in query optimization

Edge Cases¶

Null handling: Returns null if the input decimal is null (null-intolerant behavior)
Overflow behavior: Assumes the decimal's unscaled value fits within a Long range
No validation: Does not perform type checking since it's optimizer-generated
Internal use only: Not accessible through user-facing APIs

Code Generation¶

This expression fully supports Tungsten code generation through the doGenCode method, which generates efficient Java code that directly calls toUnscaledLong() on the decimal value.

Examples¶

-- Not directly available in SQL
-- Generated internally by optimizer for decimal operations

// Not directly available in DataFrame API
// Automatically created during query optimization
// Example of when it might be generated internally:
import org.apache.spark.sql.types._
val df = spark.range(1).select(lit(BigDecimal("123.45")).cast(DecimalType(5,2)))
// UnscaledValue expression may be generated internally during optimization