Update 2019-11-04: As of jython 2.7.2b2, a new artifact org.python:jython-slim now exists with proper modular dependency structure, which makes this jython-shaded project obsolete.
This project provides a rebundled version of the Jython library (specifically: org.python:jython-standalone
) with shaded (i.e., renamed) dependencies.
You can use it as follows:
<dependency>
<groupId>org.scijava</groupId>
<artifactId>jython-shaded</artifactId>
<version>2.7.1.1</version>
</dependency>
With this approach, dependencies of Jython are renamed to have a package prefix of org.scijava.jython.shaded
. So all of Jython "just works" without interfering with other consumers of JFFI or other dependencies you might have on your classpath.
It leverages the Maven Shade Plugin to do it.
Background
The Jython project is awesome. But it has a critical problem: the binary artifacts available from Maven Central are unshaded uberjars, and hence not usable in large projects. In particular, both Jython and JRuby depend on the JFFI library, but at different and incompatible versions thereof.
The naive way to add a Jython dependency is:
<dependency>
<groupId>org.python</groupId>
<artifactId>jython</artifactId>
<version>2.7.1</version>
</dependency>
But that artifact does not include the critical Lib/
directory with all of the native Python libraries necessary for many functions to work (e.g., os.path.isdir
).
So another artifact is available which includes those libraries:
<dependency>
<groupId>org.python</groupId>
<artifactId>jython-standalone</artifactId>
<version>2.7.1</version>
</dependency>
But both jython
and jython-standalone
are uberjars, meaning they bundle the dependencies of the project inside the JAR itself. And they are both unshaded, meaning they don't rename the package prefixes of those dependencies. So if you need to use a different version of those dependencies (e.g., because JRuby needs a different version of JFFI), then you are out of luck.
Example
Here is an example Jython script that produces an exception when attempting to run it using Jython 2.5.3 when JRuby 1.7.12 and dependencies (including JFFI 1.2.7) are also on the classpath:
import os
print os.path.isdir('/')
The exception which occurs is:
Traceback (most recent call last):
File "test.py", line 2, in <module>
File ".../jython-standalone-2.5.3.jar/Lib/posixpath.py", line 195, in isdir
File ".../jython-standalone-2.5.3.jar/Lib/posixpath.py", line 195, in isdir
java.lang.IncompatibleClassChangeError: Found class com.kenai.jffi.InvocationBuffer, but interface was expected
at com.kenai.jaffl.provider.jffi.AsmRuntime.marshal(AsmRuntime.java:167)
at org.python.posix.LibC$jaffl$0.stat$raw(Unknown Source)
at org.python.posix.LibC$jaffl$0.stat(Unknown Source)
at org.python.posix.BaseNativePOSIX.stat(BaseNativePOSIX.java:200)
at org.python.posix.LazyPOSIX.stat(LazyPOSIX.java:207)
at org.python.modules.posix.PosixModule$StatFunction.__call__(PosixModule.java:954)
at org.python.core.PyObject.__call__(PyObject.java:391)
at posixpath$py.isdir$17(/Applications/Science/Fiji.app/jars/jython-standalone-2.5.3.jar/Lib/posixpath.py:198)
at posixpath$py.call_function(/Applications/Science/Fiji.app/jars/jython-standalone-2.5.3.jar/Lib/posixpath.py)
at org.python.core.PyTableCode.call(PyTableCode.java:165)
at org.python.core.PyBaseCode.call(PyBaseCode.java:134)
at org.python.core.PyFunction.__call__(PyFunction.java:317)
at org.python.pycode._pyx0.f$0(New_.py:2)
at org.python.pycode._pyx0.call_function(New_.py)
at org.python.core.PyTableCode.call(PyTableCode.java:165)
at org.python.core.PyCode.call(PyCode.java:18)
at org.python.core.Py.runCode(Py.java:1275)
...
When running with jython-shaded
on the classpath instead of jython-standalone
, the above script works as expected.
FAQ
Q: Why is this structured as a multi-module project?
A: Due to a bug in ASM (the library that maven-shade-plugin
uses to rewrite the Java bytecode with the shaded package names), many of the <foo>$py.class
files cannot be processed properly. Fortunately, these classes do not actually need to be rewritten, since they themselves do not directly reference any of Jython's dependencies. So this project works around the issue by using the Shade plugin twice: once to build a JAR of shaded dependencies without the <foo>.py
classes, and a second time to mix them back in to the final artifact.