How to write cross-interpreter Python programs
==============================================

Summary:

This talk will cover basics about writing cross-interpreter
python programs. What to do and most of all what not to do.
This will help you if you want at some point in time to
run program on for example Java platform or faster python
interpreter, but also if you want to keep it running between
CPython releases.

Description:

The talk will consist mostly out of list of issues what to do and what
not to do. Some examples are given below:

Having good tests is the most important component to successful
cross-interpreter programs.  This includes having the infrastructure set up to
automatically run them on all supported platforms, runtimes, and versions.

Try to avoid 3rd party libraries that require C extension modules unless you can
fallback to a pure Python replacement.  Implementation specific code, especially
if it's not Python code, is one of the hardest parts to port to other
interpreters.  This also applies to libraries that are made available by the
implementation by default but are not part of the stdlib like Java libraries
from Jython and .NET in IronPython.  The preferred way to access libraries is
ctypes, since CPython, PyPy and IronPython supports it,
while Jython is on the way.

Mind your bytes and strings!  The usual rule is to always decode data as soon as
possible from IO if you need to do processing on it, or always keep it as bytes.
In 2.x, always use unicode for text, and str for byte strings.

Don't use sys.prefix.  Different implementations have different installed
layouts, so it's wrong to assume the CPython one.

Use all new style classes.  Not only does this provide consistency, but it will
avoid surprises in 3.x when all classes are new style and be useful for other
implementations, where new style classes are optimized more.

Never use exact string representation of exceptions, they might even
change from one minor revision to other.

Don't really on exact finalization, especially on files.

In short: there is no feature obscure enough for people not to rely on.