Fun with relative imports and py3k

[Kay Schluehr]

Since their introduction in Python 2.5 I only reviewed the new
"relative import" notation briefly by reading the "What's new in
Python 2.5" article. Now I wanted checkout if I get comfortable with
them.

Opening the tutorial I found following notice ( 6.4.2 ):

"Note that both explicit and implicit relative imports are based on
the name of the current module.
Since the name of the main module is always "__main__", modules
intended for use as the main module
of a Python application should always use absolute imports."

The distinction between modules and scripts was new to me. One of the
primary features I always appreciated in Python was that modules had a
dual purpose: they were components and they were programs. One could
use them in isolation or as a part of a package something I always
missed when I programmed in languages with the tyranny of a single
main.

However 'implicit relative imports' work as usual and Py3K just
changes the priority of search - one might simply ignore the
'explicit' variant and care for unambiguous names.

Now things are not so easy in life and when you are confronted with
other peoples software you have to know at least the rules of the
game.

Digging a little deeper I found the following section in "What's new
in Python 2.6"

'Python’s -m switch allows running a module as a script.
When you ran a module that was located inside a package, relative
imports didn’t work correctly.'

Hmm... they worked as specified or at least as documented. I admit I'm
confused but I suspect I'm not the only one. Now everything is fine.
You just run each and every module with the -m option "as a script"
because there is a chance that somewhere is a relative import ( e.g.
local to a function ) and suddenly the module fails to import stuff
due to a modality in the import behavior.

Let's see how it works in practice.

This here my very first attempt to start Pythons 2to3 refactoring tool
written by the master himself. I looked at it and he used relative
imports.

C:\Python30\Lib \lib2to3>C:\Pyt hon30\python -m refactor.py
Traceback (most recent call last):
File "C:\Python30\li b\runpy.py", line 103, in _run_module_as_ main
loader, code, fname = _get_module_det ails(mod_name)
File "C:\Python30\li b\runpy.py", line 78, in _get_module_det ails
loader = get_loader(mod_ name)
File "C:\Python30\li b\pkgutil.py", line 456, in get_loader
return find_loader(ful lname)
File "C:\Python30\li b\pkgutil.py", line 466, in find_loader
for importer in iter_importers( fullname):
File "C:\Python30\li b\pkgutil.py", line 422, in iter_importers
__import__(pkg)
File "C:\Python30\li b\lib2to3\refac tor.py", line 23, in <module>
from .pgen2 import driver
ValueError: Attempted relative import in non-package
Not working. I'm perfectly sure I don't understand the error message
but what's worse is that runpy.py fails which
was just created ( in Python 2.6 ) to fix the issue with the
apparently broken relative imports that is o.k. according to the
introductory material that is dedicated to newbies.

I'm trying to fix this naively by turning the explicit into implicit
relative (or absolute?) imports and see what happens.

refactor.py
------------------------
...

# Local imports
from pgen2 import driver
from pgen2 import tokenize

import pytree
import patcomp
import fixes
import pygram
....
C:\Python30\Lib \lib2to3>C:\Pyt hon30\python -m refactor.py
Traceback (most recent call last):
File "C:\Python30\li b\runpy.py", line 103, in _run_module_as_ main
loader, code, fname = _get_module_det ails(mod_name)
File "C:\Python30\li b\runpy.py", line 78, in _get_module_det ails
loader = get_loader(mod_ name)
File "C:\Python30\li b\pkgutil.py", line 456, in get_loader
return find_loader(ful lname)
File "C:\Python30\li b\pkgutil.py", line 466, in find_loader
for importer in iter_importers( fullname):
File "C:\Python30\li b\pkgutil.py", line 422, in iter_importers
__import__(pkg)
File "refactor.p y", line 27, in <module>
import patcomp
File "C:\Python30\li b\lib2to3\patco mp.py", line 17, in <module>
from .pgen2 import driver
ValueError: Attempted relative import in non-package

This time patcomp fails with its relative imports despite the fact
that it is not imported as a "script" but as a module inside the
lib2to3 package and it is not __main__ but refactor.py is.

I give up. I always thought migration from Python 2.X to Python 3.0
was an issue not the conversion tool ;)