Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
95 lines (60 sloc) 2.99 KB

PyJVM implementation


PyJVM is python 2.7 based implementation of Java Virtual Machine 7.

It's loosely based on jdk7 specification (see oracle's web site). Most exceptions are in runtime checks and binary compatibility. For example, if one's try to call other's class private method - real jdk will throw an exception, while pyjvm will execute it. Basically, java has both compile and run time, and pyjvm has none - it's assuming that source of java classes is trusted.

PyJVM supports all byte code operations and has very few python coded shortcuts.

Project structure (under pyjvm):

  • ops: all java byte codes
  • platform: all native method calls (see
  • *.py: see comments in those files


All java types are supported.

int, byte, short, char are plain python int.

float, double and long are tuples e.g. ("long", 123)

References - to instances in heap - are also tuples ("ref", 34). None for null.

VM owned objects: ("vm_ref", -1)

Classes are loaded with class loader from folders or jars and are represented by JavaClass py object (also java.lang.Class instance may be created).

VM initialization

For every fresh pyjvm instance there is an initialization process. This process is responsible for loading all initial dependencies and creating system wide objects - for example, initializing System.out. See init for details.

To speed up this phase vm caching is implemented. For every fresh start, after vm is initialized and is about to run user application, file vm-cache.bin is created with memory dump.

On every start, pyjvm checks if vm-cache.bin exists and if it's version matches current code base. If it is - vm is loaded from the file.

There is constant: SERIALIZATION_ID in - it should be incremented every time when init process get changed. To reset caching for those users who has old version already cached.

To ignore caching process pyjvm can be run with parameter: -novmcache

Memory structure


perm_gen - storage for py objects for java classes. Name based dictionary. When code requests a specific class to get loaded this cache is always checked first. There are no concept of class + classloader relation in pyjvm. Overall, this should be refactored at some point.

heap - java heap. Number based dictionary; so a reference to an object is ("ref", 4). Where number is the key. Heap contains only instance fields, byte code itself (as well as static fields) is in perm_gen.

Garbage collection

See for details


Multithreading and synchronization is supported. Based on "one-core" machine, so all threads are executed in order based on small quota. See


When java exception is about to be thrown - py instance of JavaException is created. This JE instance has a filed - ref - which is a reference to actual exception (e.g. java.lang.NullPointerException) object created in heap.