demain 0.9

Background

When you run a Python script from the command line, or invoke a module using the -m option, Python gives that script or module the fake name '__main__' so that it can detect that it is running as the “main program.” Consider a small script that looks like this:

# foo.py
print __name__

Invoking it from the command line obliterates the normal foo name which it would have otherwise (as we can demonstrate, by importing the module from the command line):

$ python foo.py
__main__
$ python -m foo
__main__
$ python -c 'import foo'
foo

This name mangling causes no end of trouble when used with modules that themselves are targets of importation, because the first attempt to import the module under its real name results in a duplicate of the module being created, with its own copy of every object and class. The bugs that this causes tend to be particularly difficult to track down.

The demain.fix() function is designed to prevent this problem from even starting, by being the very first thing that you call at the top of every module that you intend to be invoked directly. Be careful not only to avoid doing other imports before before running fix() but also make sure that you create no classes first; otherwise, the name '__main__' will get copied into your classes as their module’s name.

I knew that I finally had to do something about this problem when I read Defending Pyramid’s Design and reached Chris McDonough’s anguished plea in the section titled, “Application Programmers Don’t Control The Module-Scope Codepath.”

http://docs.pylonshq.com/pyramid/dev/designdefense.html

There, his long experience with Python and his sharply-honed coding practices lead him to make some excellent recommendations against the dangers that small Python applications — and in particular those written using Web micro-frameworks — incur when they treat a script as though it were a normal module. My hope is that demain will solve this problem by making an invoked Python script a safe place for objects, instead of having them all duplicated when the same Python file is imported from elsewhere in the application using a normal import.

This package comes with a modest test suite, built from a few hand-built examples, as well as McDonough’s sample decorator framework that shows the problem with Python’s handling of script invocation. To run it, simply type:

$ python -m demain.tests