Interfacing From Non-Java Code

Software written in Java is easiest to use with other Java code. However, it is possible to call Java code from a program written in another language. But how to do so depends on your program's needs.

Technologically, there are two broad categories of solutions: in-process approaches, and inter-process communication. See below for discussion and examples of each.

For further reading, check out Codemesh's technology comparison.

In-process approaches

With an in-process approach, your application directly invokes Java code, either by spawning its own internal Java Virtual Machine (JVM) and passing data across a bridge, or otherwise executing the Java code within a single environment.

StrengthsWeaknesses
  1. Tight (API-level) integration
  2. Minimal performance overhead
  3. Few security considerations
  1. No shared state between processes
  2. Limited portability

Paradigms:

  • JNI – The most common paradigm is the Java Native Interface, an API for interfacing Java programs with native C/C++ code. Functionality exists to 1) call C/C++ methods from Java, and 2) spawn a Java Virtual Machine and execute Java instructions from C/C++. The latter direction, known as Java Invocation, is the relevant one for calling Java code from C++. Because using the JNI directly involves writing a lot of tedious glue code, several projects have emerged for autogenerating such code. See below: raw JNI, Jace, JunC++ion, JuggerNET
  • Compilers – The dominant paradigm in Java is to compile Java source to Java bytecode, then execute the bytecode in a Java interpreter. However, an alternative is to compile the Java source directly to native code so that it can link with other native programs. Such an approach requires that the compiler provide correct support for all necessary Java standard library features. It may also exhibit much different performance (for better or for worse) than Sun's Java implementation does. See below: GCJ
  • Runtimes – The safest way to guarantee correct program behavior is to execute Java bytecode using the Java interpreter(s) with which it has been tested. However, a Java runtime written using a specific framework (e.g., .NET) could enable seamless integration with other (non-Java) programs within the same framework. Like the compiler-based paradigm above, though, it is reliant on the correctness, completeness and performance of the Java implementation in question. See below: IKVM.NET

Inter-process communication

The other approach is inter-process communication, a broad collection of techniques for exchanging data between multiple running programs. Such techniques take many forms in computing; one ubiquitous example is web browsers and web servers. Most solutions in this category are some form of middleware.

StrengthsWeaknesses
  1. Share state between multiple processes on multiple machines
  2. Broad portability and language support
  1. Object marshalling incurs significant overhead
  2. Potentially vulnerable to security exploits

Paradigms:

  • Local communication – Modern operating systems provide several ways to share information between processes, including shared memory, file system access, and passing data back and forth with standard input and output streamsSee below: pipes, files
  • Messaging – Networking technology allows a process on one computer to send and receive messages from another process on a different machine. The client-server model is probably most applicable for Java/native integration, with the Java portion acting as a server that can be queried from the native code. See below: sockets, XML-RPC
  • ORB – An object request broker (ORB) is a high-level form of middleware for transferring objects between multiple running programs. ORBs provide an abstraction that can reduce and simplify code written by providing access to a wealth of higher-level messaging features. See below: Ice, CORBA, Codemesh Shared JVM

List of solutions

It is a significant challenge to access a complex Java API from code written in another language, especially in a cross-platform and high performance way. The table below provides an overview of viable approaches, with links to source code and instructions where appropriate. Which approach to use depends on your application's target platforms and languages, and the interaction model between your application and the Java code—see the discussion of in-process solutions versus inter-process communication above for details.

A note about SWIG. The Simplified Wrapper and Interface Generator (SWIG) is an excellent tool for exposing C++ functionality to higher level languages such as Java. Unfortunately, calling native code from Java is the wrong direction for our purposes. However, when combined with an integration solution specific to C++, SWIG could be used to extend that solution into other languages (see SWIG's list of supported languages for a complete list).

SolutionTypeLanguagesNotes
Raw JNIIn-process (JNI)C/C++
  • You can code your integration layer using pure JNI calls (but we don't recommend it).
  • Low-level JNI offers full control over the interface between Java and native code.
  • Raw JNI solutions are time-consuming and error-prone to implement.
  • We have coded a simple example for calling the Bio-Formats library in this way.
  • We recommend a higher level integration solution such as Jace instead.
JaceIn-process (JNI)C/C++
  • Jace generates C++ proxy classes, one per Java class, mirroring the original functionality as much as possible.
  • The C++ proxy classes use JNI under the hood but handle most of the usual JNI pitfalls.
  • We provide a set of Bio-Formats C++ bindings using Jace.
JunC++ionIn-process (JNI)C/C++
  • JunC++ion is a commercial in-process integration solution for C/C++ available from Codemesh, Inc.
JuggerNETIn-process (JNI).NET
  • JuggerNET is a commercial in-process integration solution for .NET available from Codemesh, Inc.
GCJIn-process (compiler)C/C++ (GCC only)
  • GCJ can compile Java code into machine code.
  • Instead of JNI, GCJ uses its Compiler Native Interface (CNI) to enable access to Java code from C++.
  • Works with GCC only (i.e., not with Microsoft Visual C++ or other compilers).
  • Correctness and performance is dependent on the compiler implementation.
IKVM.NETIn-process (runtime).NET/Mono
  • IKVM.NET interprets Java byte code on the fly into the .NET/Mono framework.
  • Interoperability is limited to applications in the .NET/Mono framework.
  • Correctness and performance is dependent on the runtime implementation.
PipesInter-process (local)Any
  • Pipes can work well when quantity of data being transferred is limited.
  • Be careful about blocking operations causing deadlock.
  • The OME Perl server uses a combination of pipes and files to interface with the Bio-Formats library.
FilesInter-process (local)Any
  • Communication via files is slower than pipes, since messages go through disk.
  • The size of the communication buffer is limited only by available disk space.
  • The OME Perl server uses a combination of pipes and files to interface with the Bio-Formats library.
SocketsInter-process (messaging)Any
  • You can use a sockets API directly to create a custom solution (but we don't recommend it).
  • JVMLink was our first cut at such a solution, before we realized that we were essentially inventing our own middleware.
  • We recommend using Ice or CORBA instead.
XML-RPCInter-process (messaging)Many
  • XML-RPC is a cross-platform remote procedure call (RPC) technology using XML.
  • XML-RPC is a human readable, but inefficient, means of transporting information.
  • See also: SOAP
IceInter-process (ORB)Several
  • Ice is high-performance middleware for cross-platform client/server communication.
  • ZeroC argues that Ice is superior to CORBA.
  • The OMERO server uses Ice extensively to provide client/server interoperability cross-language.
CORBAInter-process (ORB)Many
  • Java RMI over IIOP enables communication between Java and other CORBA-supported languages.
  • CORBA is a mature technology with widespread use in the enterprise community.
  • Support for CORBA is built in to the Java 2 platform.
  • Nonetheless, where possible we suggest you use Ice instead.
Codemesh Shared JVMInter-process (ORB)C++, .NET
  • The Codemesh Shared JVM is a commercial inter-process integration solution available from Codemesh, Inc.

Know a great integration solution that we missed? Let us know!