The main Python implementation is written in the C language and is called CPython. It is the one that the majority of people refer to when they talk about Python. When the language evolves, the C implementation is changed accordingly. Besides C, Python is available in a few other implementations that are trying to keep up with the mainstream. Most of them are a few milestones behind CPython, but provide a great opportunity to use and promote the language in a specific environment.
There are plenty of alternative Python implementations available. The Python Wiki page on that topic (https://wiki.python.org/moin/PythonImplementations) features more than 20 different language variants, dialects, or implementations of Python interpreter built with something else than C. Some of them implement only a subset of the core language syntax, features, and built-in extensions but there is at least a few that are almost fully compatible with CPython. The most important thing to know is that while some of them are just toy projects or experiments, most of them were created to solve some real problems – problems that were either impossible to solve with CPython or required too much of the developer's effort. Examples of such problems are:
Running Python code on embedded systems
Integration with code written for runtime frameworks such as Java or .NET or in different languages
Running Python code in web browsers
This section provides a short description of subjectively most popular and up-to-date choices that are currently available for Python programmers.
Stackless Python advertises itself as an enhanced version of Python. Stackless is named so because it avoids depending on the C call stack for its own stack. It is in fact a modified CPython code that also adds some new features that were missing from core Python implementation at the time Stackless was created. The most important of them are microthreads managed by the interpreter as a cheap and lightweight alternative to ordinary threads that must depend on system kernel context switching and tasks scheduling.
The latest available versions are 2.7.9 and 3.3.5 that implement 2.7 and 3.3 versions of Python respectively. All the additional features provided by Stackless are exposed as a framework within this distribution through the built-in stackless module.
Stackless isn't the most popular alternative implementation of Python, but it is worth knowing because ideas introduced in it have had a strong impact on the language community. The core switching functionality was extracted from Stackless and published as an independent package named greenlet, which is now a basis for many useful libraries and frameworks. Also, most of its features were re-implemented in PyPy—another Python implementation that will be featured later. Refer to http://stackless.readthedocs.org/.
Jython is a Java implementation of the language. It compiles the code into Java byte code, and allows the developers to seamlessly use Java classes within their Python modules. Jython allows people to use Python as the top-level scripting language on complex application systems, for example, J2EE. It also brings Java applications into the Python world. Making Apache Jackrabbit (which is a document repository API based on JCR; see http://jackrabbit.apache.org) available in a Python program is a good example of what Jython allows.
The latest available version of Jython is Jython 2.7, and this corresponds to 2.7 version of the language. It is advertised as implementing nearly all of the core Python standard library and uses the same regression test suite. The version of Jython 3.x is under development.
The main differences of Jython as compared to CPython implementation are:
The main weakness of this implementation of the language is the lack of support for C Python Extension APIs, so no Python extensions written in C will work with Jython. This might change in the future because there are plans to support the C Python Extension API in Jython 3.x.
Some Python web frameworks such as Pylons were known to be boosting Jython development to make it available in the Java world. Refer to http://www.jython.org.
IronPython brings Python into the .NET Framework. The project is supported by Microsoft, where IronPython's lead developers work. It is quite an important implementation for the promotion of a language. Besides Java, the .NET community is one of the biggest developer communities out there. It is also worth noting that Microsoft provides a set of free development tools that turn Visual Studio into full-fledged Python IDE. This is distributed as Visual Studio plugins named PVTS (Python Tools for Visual Studio) and is available as open source code on GitHub (http://microsoft.github.io/PTVS).
The latest stable release is 2.7.5 and it is compatible with Python 2.7. Similar to Jython, there is some development around Python 3.x implementation, but there is no stable release available yet. Despite the fact that .NET runs primarily on Microsoft Windows, it is possible to run IronPython also on Mac OS X and Linux. This is thanks to Mono, a cross platform, open source .NET implementation.
Main differences or advantages of IronPython as compared to CPython are as follows:
Similar to Jython, the lack of GIL (global interpreter lock) allows the better utilization of multiple cores in multi-threaded applications
Code written in C# and other .NET languages can be easily integrated in IronPython and vice versa
Can be run in all major web browsers through Silverlight
When speaking about weaknesses, IronPython, again, seems very similar to Jython because it does not support the C Python Extension APIs. This is important for developers who would like to use packages such as numpy that are largely based on C extensions. There is a project called ironclad (refer to https://github.com/IronLanguages/ironclad) that aims to allow using such extensions seamlessly with IronPython, albeit its last known supported release is 2.6 and development seems to have stopped at this point. Refer to http://ironpython.net/.
PyPy is probably the most exciting implementation, as its goal is to rewrite Python into Python. In PyPy, the Python interpreter is itself written in Python. We have a C code layer carrying out the nuts-and-bolts work for the CPython implementation of Python. However, in the PyPy implementation, this C code layer is rewritten in pure Python.
This means you can change the interpreter's behavior during execution time and implement code patterns that couldn't be easily done in CPython.
PyPy currently aims to be fully compatible with Python 2.7, while PyPy3 is compatible with Python 3.2.5 version.
In the past, PyPy was interesting mostly for theoretical reasons, and it interested those who enjoyed going deep into the details of the language. It was not generally used in production, but this has changed through the years. Nowadays, many benchmarks show that surprisingly PyPy is often way faster than the CPython implementation. This project has its own benchmarking site that tracks the performance of each version measured using tens of different benchmarks (refer to http://speed.pypy.org/). It clearly shows that PyPy with JIT enabled is at least a few times faster than CPython. This and other features of PyPy makes more and more developers decide to switch to PyPy in their production environments.
The main differences of PyPy as compared to the CPython implementation are:
Garbage collection is used instead of reference counting
Integrated tracing JIT compiler that allows impressive improvements in performance
Application-level Stackless features are borrowed from Stackless Python
Like almost every other alternative Python implementation, PyPy lacks the full official support of C Python Extension API. Still it, at least, provides some sort of support for C extensions through its CPyExt subsystem, although it is poorly documented and still not feature complete. Also, there is an ongoing effort within the community in porting NumPy to PyPy because it is the most requested feature. Refer to http://pypy.org.