DateSub¶

Overview¶

The DateSub expression subtracts a specified number of days from a given date. It operates on date values and returns a new date that is the specified number of days earlier than the input date.

Syntax¶

date_sub(start_date, num_days)

Arguments¶

Argument	Type	Description
startDate	DateType	The starting date from which to subtract days
days	IntegerType, ShortType, ByteType	The number of days to subtract from the start date

Return Type¶

DateType - Returns a date value representing the result of the subtraction operation.

Supported Data Types¶

Input date: DateType only
Days parameter: IntegerType, ShortType, or ByteType (any integer numeric type in the TypeCollection)

Algorithm¶

Converts the start date to its internal integer representation (days since epoch)
Converts the days parameter to an integer value using intValue()
Performs integer subtraction: startDate - days
Returns the result as a date value
Uses null-safe evaluation to handle null inputs properly

Partitioning Behavior¶

This expression does not affect partitioning behavior:

Preserves existing partitioning as it operates on individual rows
Does not require shuffle operations
Can be pushed down to individual partitions for parallel execution

Edge Cases¶

Null handling: Returns null if either input parameter is null (null-intolerant behavior)
Negative days: Subtracting negative days effectively adds days to the date
Integer overflow: Uses standard integer arithmetic, so extremely large day values may cause overflow
Date boundaries: Results outside valid date ranges may cause issues depending on the underlying date implementation

Code Generation¶

This expression supports Tungsten code generation through the doGenCode method, which generates optimized Java code for the subtraction operation: ${ev.value} = $sd - $d;

Examples¶

-- Subtract 5 days from a specific date
SELECT date_sub('2016-07-30', 5);
-- Result: 2016-07-25

-- Subtract 1 day from a date column
SELECT date_sub(order_date, 1) FROM orders;

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

df.select(date_sub(col("start_date"), lit(7)))
df.withColumn("previous_week", date_sub(col("current_date"), 7))