Abs¶

Overview¶

The Abs expression computes the absolute value of a numeric expression or ANSI interval. It returns the magnitude of the input value without regard to its sign, effectively removing any negative sign from the input.

Syntax¶

ABS(expr)

// DataFrame API
col("column_name").abs
abs(col("column_name"))

Arguments¶

Argument	Type	Description
child	Expression	The input expression to compute the absolute value of
failOnError	Boolean	Whether to fail on arithmetic overflow (defaults to ANSI mode setting)

Return Type¶

Returns the same data type as the input expression. The output type exactly matches the input type for all supported numeric types and interval types.

Supported Data Types¶

All numeric types: ByteType, ShortType, IntegerType, LongType, FloatType, DoubleType
DecimalType with any precision and scale
DayTimeIntervalType (using LongExactNumeric)
YearMonthIntervalType (using IntegerExactNumeric)

Algorithm¶

For decimal types: Uses the native abs() method of the decimal representation
For integer types in ANSI mode: Uses MathUtils.negateExact() to handle overflow detection when negating negative values
For ANSI interval types: Uses MathUtils.negateExact() with overflow protection
For other numeric types: Uses java.lang.Math.abs() with appropriate type casting
Leverages type-specific Numeric instances for evaluation in interpreted mode

Partitioning Behavior¶

This expression preserves partitioning as it is a deterministic unary transformation:

Does not require shuffle operations
Maintains the same partitioning scheme as the input
Can be pushed down to individual partitions independently

Edge Cases¶

Null handling: Returns null for null input (null intolerant behavior)
Overflow behavior: In ANSI mode (failOnError = true), throws an exception on integer overflow (e.g., abs(Long.MinValue))
Non-ANSI mode: May silently overflow for integer edge cases like Long.MinValue
Decimal precision: Maintains the same precision and scale as input, no precision loss
Floating point: Handles NaN, positive/negative infinity according to java.lang.Math.abs() semantics

Code Generation¶

This expression supports Tungsten code generation with optimized paths:

Uses native decimal abs() method for decimal types
Generates overflow-safe code using MathUtils.negateExact() for ANSI integer and interval types
Falls back to java.lang.Math.abs() for standard numeric types
Includes proper type casting in generated code

Examples¶

-- Numeric examples
SELECT ABS(-42);           -- Returns: 42
SELECT ABS(3.14);          -- Returns: 3.14
SELECT ABS(-3.14);         -- Returns: 3.14

-- Interval examples
SELECT ABS(INTERVAL -'1-1' YEAR TO MONTH);  -- Returns: 1-1
SELECT ABS(INTERVAL -'2 3:4:5' DAY TO SECOND); -- Returns: 2 03:04:05

-- Null handling
SELECT ABS(NULL);          -- Returns: NULL

// DataFrame API examples
import org.apache.spark.sql.functions._

df.select(abs(col("amount")))
df.select(col("value").abs)
df.withColumn("abs_difference", abs(col("value1") - col("value2")))