BinaryPad¶

Overview¶

BinaryPad is a runtime-replaceable expression that provides left padding (lpad) and right padding (rpad) functionality for binary data types. It serves as a wrapper that delegates the actual padding logic to native methods in the ByteArray class, supporting padding of binary strings with a specified pad value to reach a target length.

Syntax¶

LPAD(binary_expr, length, pad_binary)
RPAD(binary_expr, length, pad_binary)

// DataFrame API usage would depend on the registered function name
col("binary_column").expr("lpad(binary_column, 10, pad_bytes)")

Arguments¶

Argument	Type	Description
str	BinaryType	The binary expression to be padded
len	IntegerType	Target length for the padded result
pad	BinaryType	Binary value used for padding

Return Type¶

BinaryType - Returns a binary array with the specified padding applied.

Supported Data Types¶

Input: BinaryType for string and pad arguments, IntegerType for length
Output: BinaryType
Implicit casting is supported for input types as this expression extends ImplicitCastInputTypes

Algorithm¶

Validates that funcName is either "lpad" or "rpad" through assertion
Creates a StaticInvoke expression targeting the ByteArray class
Delegates actual padding computation to native ByteArray.lpad() or ByteArray.rpad() methods
Maintains input type constraints through the inputTypes override
Uses runtime replacement pattern to substitute the expression during query planning

Partitioning Behavior¶

This expression preserves partitioning characteristics:

Does not require shuffle operations
Can be applied per-partition independently
Maintains data locality since it operates on individual rows

Edge Cases¶

Null handling: Returns null if any input argument is null (returnNullable = false in StaticInvoke, but null propagation handled by input validation)
Empty binary input: Behavior depends on underlying ByteArray implementation
Zero or negative length: Handled by the native ByteArray methods
Empty pad value: May result in truncation or error depending on implementation
Length exceeding limits: Subject to ByteArray implementation constraints

Code Generation¶

This expression uses StaticInvoke for code generation, which enables Tungsten code generation by calling native methods directly rather than using interpreted evaluation mode.

Examples¶

-- Left pad binary data to 8 bytes with zero bytes
SELECT LPAD(CAST('ABC' AS BINARY), 8, CAST('\x00' AS BINARY));

-- Right pad binary data to 10 bytes  
SELECT RPAD(binary_column, 10, CAST('XY' AS BINARY)) FROM table;

// DataFrame API usage (assuming registered functions)
import org.apache.spark.sql.functions._
df.select(expr("lpad(binary_col, 8, cast('\\x00' as binary))"))
df.select(expr("rpad(binary_col, 10, cast('pad' as binary))"))