13  Appendix: Manual Bytecode Compilation

This chapter provides a conceptual walk-through of how R bytecode assembly is turned into a bytecode object.

13.1 Example bytecode assembly

The example bytecode assembly is simply performing the addition of 2 numbers.

LDCONST 10
LDCONST 20
ADD
RETURN

13.2 Initialise consts list

Storage is first created for the constants which appear in the code. Note that this storage uses 0-based indexing as it is referenced from C, not R.

Initialising consts storage

13.3 Store constants

Constants and expressions are stored in the consts area.

In the bytecode, it is the index to the storage location that is the actual argument to the instruction.

Storing values in consts

LDCONST <0>
LDCONST <1>
ADD
RETURN

13.4 Stored expressions

The ADD instruction as defined in R requires a reference to a stored expression. In {rbytecode} this is automatically created as a dummy quoted expression that is helpful when debugging.

This stored expression is placed into consts and the argument to the ADD instruction is the index of this storage location.

Storing expressions in consts

LDCONST <0>
LDCONST <1>
ADD     <2>
RETURN

13.5 Lookup integer code for each instruction.

The string representing the instruction is turned into an integer, based upon its location in the full list of instruction codes stored in compiler:::Opcode.names.

head(compiler:::Opcodes.names)

[1] "BCMISMATCH.OP" "RETURN.OP"     "GOTO.OP"       "BRIFNOT.OP"   
[5] "POP.OP"        "DUP.OP"       

The opcode for each instruction in the example bytecode assembly is determined and takes the place of the string representation.

match(c('LDCONST.OP', 'ADD.OP', 'RETURN.OP'), compiler:::Opcodes.names) - 1L

[1] 16 44  1

Final state of consts storage

<16> <0>
<16> <1>
<44> <2>
<1>

13.6 Final bytecode.

The almost final bytecode:

1. vector of integers: c(16, 0, 16, 1, 44, 2, 1)
2. list of consts: list(10, 20, quote(stop("ADD")))

To prevent inconsistencies in code execution when the bytecode instruction set is updated, the vector of bytecode integers is prefixed with the bytecode version number. In R v4.3.1 the version number is 12.

The final bytecode object is then:

1. vector of integers: c(12, 16, 0, 16, 1, 44, 2, 1)
2. list of consts: list(10, 20, quote(stop("ADD")))