Here's a usage where this matters.  It's a simplification of the bug report that I got that prompted me to open this.  Let's say I have the following code:

/usr/lib/python2.7/site-packages/foo.py::

    def help():
        """I'm a little docstring, short and sweet"""
        print help.__doc__

/usr/bin/bar.py::
    #!/usr/bin/python -tt
    import sys
    import foo

    if "--help" in sys.argv:
        foo.help()
    else:
        print "if you type --help, it's me you'll meet"

The system administrator on this box comes along and runs::

    $ PYTHONOPTIMIZE=2 bar.py
    if you type --help, it's me you'll meet

No problems apparent there but then, the user comes along later and runs::
    $ PYTHONOPTIMIZE=1 ./bar.py --help
    None

At this point the end user opens a bug against my software telling me that --help is broken which confuses everyone.

Not sure the best way to fix this -- ideas that pop into my head:

These solutions don't lead to a bug report as python does the right thing on its own:

* python reads and write separate pyo files for the different optimization levels (foo.pyo1 foo.pyo2)
* python autodetects whether the cached .pyo was written for the current optimization level and disregards its existence if there's a mismatch (leading to rewriting if the user has permissions, otherwise, recompilation but no writing to disk).

The following solution leads to a bug report but can be diagnosed with just the bug reporter.  Note that PYTHONOPTIMIZE= /usr/bin/bar.py is almost as good in this situation:

* python has a commandline switch to disregard current .pyo files.

These solutions lead to a bug report but can be diagnosed with the cooperation of the system administrator in addition to the bug reporter:

* Command line switch that overwrites the .pyo files
* rm -rf *.pyo