From source to bytecode

Key points

• javac checks types and writes a .class file. It optimises almost nothing - that job belongs to the JVM at runtime.
• Bytecode is a stack machine. Push the operands, run one instruction, it pops them and pushes the result.
• There are around 200 opcodes and most are dull: iload, iadd, invokevirtual, areturn.
• javap -c disassembles any class file. It is the fastest way to answer "what did the compiler actually do here".
• Every class file starts with the magic number 0xCAFEBABE and a version. Class file 65 means Java 21.
• Run a newer class file on an older JVM and you get UnsupportedClassVersionError. Bytecode is forward compatible, not backward.

Example

# compile, then look at what javac actually produced
javac Adder.java

javap -c Adder
#   int add(int, int);
#     0: iload_1          // push the first argument
#     1: iload_2          // push the second argument
#     2: iadd             // pop both, push the sum
#     3: ireturn          // return the top of the stack

# the header: magic number, class file version, constant pool
javap -v Adder | head -20

Bytecode is a small stack machine, and javap -c shows it to you.