.CLASS File Format

Introduction

A .class file holds one compiled class, interface, enum, record, annotation interface or module declaration. It carries the class’s methods as bytecode for the Java Virtual Machine, and every name, type and constant they use. The javac compiler of OpenJDK writes one per class, nested and anonymous classes included, and the JVM loads each when the program first needs it. Kotlin, Scala, Groovy, Clojure and other JVM languages write the same format.

This is the format chapter 4 of the Java Virtual Machine Specification describes. It began with JDK 1.0.2 in 1996 (version 45.3), and every Java release since has added to it without changing what was there: new kinds of constant, new attributes, new flags and one new instruction. A reader that knows an older version can still find its way through a newer file, because every attribute says how long it is.

The pre-release versions 45.0 to 45.2 may use a different Code attribute; files laid out that way are a separate format, the Oak class file. See all the class file variations.

A .class file dropped on any page of this site opens in the explorer, which shows every structure in it, a disassembly in the style of javap -c -v and the constant pool as a table.

Structure

A class file is one structure, read from start to end. All numbers are big-endian and unsigned unless stated. Only the first four fields have fixed offsets: the constant pool’s entries have different sizes, so everything after it has to be found by reading through.

Almost everything after the constant pool refers into it by a two-byte index: a class’s name, a method’s descriptor, the string an instruction loads. The bytecode of each method sits in its Code attribute, and its instructions name fields, methods and classes through the same indexes.

What each version added

Everything below is an addition: a file may use a feature only from the version that introduced it, and nothing earlier changed its layout.

The language features behind these are described on the Java versions page. Major versions 62 to 72 add nothing to the layout that this reference covers.

Telling it apart

The explorer’s parser for this format checks that the file starts with CA FE BA BE and that the major version, the two bytes at offset 6, is 45 or more. It then reads the whole file strictly: every constant pool entry must have a known tag, every attribute it knows must fill exactly the length it gives, every method’s bytecode must be made of real instructions, and nothing may follow the last attribute. A file that fails is turned down, and the explorer tries the Oak class file parser.

On an Oak file, this parser reads the 4-byte Code header as if it were 8 bytes long. The code_length it finds is the start of the bytecode, which runs past the end of the attribute, so the file is turned down. A version 45.0 to 45.2 file laid out as the specification says is read here.

Variants

• Java class file – versions 45.3 to 72, JDK 1.0.2 to Java SE 28, as the JVM Specification describes
• Oak class file (45.0 to 45.2) – pre-release classes whose Code attribute has a one-byte max_stack and max_locals

Links