In Python 3.3 (I did not test with 3.4), sys.modules cannot be reassigned without breaking the import mechanism; while it works with Python <= 3.2.

Here is how to reproduce the bug:

progval@Andromede:/tmp$ mkdir test_imports
progval@Andromede:/tmp$ echo "from . import module" > test_imports/__init__.py
progval@Andromede:/tmp$ echo "print('foo')" > test_imports/module.py
progval@Andromede:/tmp$ python3.3
Python 3.3.1 (default, Apr  6 2013, 13:58:40) 
[GCC 4.7.2] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.modules = dict(sys.modules)
>>> import test_imports
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "./test_imports/__init__.py", line 1, in <module>
    from . import module
SystemError: Parent module 'test_imports' not loaded, cannot perform relative import
>>> 
progval@Andromede:/tmp$ python3.2
Python 3.2.3 (default, May  6 2013, 01:46:35) 
[GCC 4.7.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.modules = dict(sys.modules)
>>> import test_imports
foo
>>>

I wouldn't call it a bug personally. The modules dictionary is used in all kinds of places in the interpreter; you can change the dictionary's contents, but not swap it with another one.

It's just a pity that we can't forbid reassignment altogether.

There is no good way to solve this. At the C level there interpreter struct has two key fields, sysdict and modules. The former is sys.__dict__ and the latter is sys.modules. But when you re-assign sys.modules you then break the assumption that sys.modules is the same dict as that contained in interp->modules. And this all goes out the window as the C code is expected to use interp->modules while the Python code in importlib only has access to sys.modules. The reason this used to "work" is your new dictionary was completely ignored and so we basically a no-op from the perspective of import (done in Python 2.7 but same result in any version up to Python 3.3)::

  >>> import sys
  >>> original_modules = sys.modules
  >>> new_modules = sys.modules.copy()
  >>> sys.modules = new_modules
  >>> import pkg
  >>> 'pkg' in original_modules
  True
  >>> 'pkg' in new_modules
  False

What really needs to happen is that sys.modules needs to be documented as something that cannot be replaced. If you really want to update it cleanly then do sys.modules.clear(); sys.modules.update(new_modules), but even that is tricky because removing certain modules will flat-out break Python.

I have updated the issue to be a documentation one and added Python 3.4 to the affected versions.

"sys.modules
This is a dictionary that maps module names to modules which have already been loaded. This can be manipulated to force reloading of modules and other tricks."

How about adding at the end
"However, replacing or clearing it may cause Python to fail."

or to be slightly more explicit "However, replacing it or deleting critical entries may cause Python to fail."

Sounds good to me.

Documentation added for sys.modules

New changeset 4f8160e45cb7 by Brett Cannon in branch '3.3':
Issue bpo-17953: document that sys.modules shouldn't be replaced (thanks
http://hg.python.org/cpython/rev/4f8160e45cb7

New changeset b60ae4ebf981 by Brett Cannon in branch 'default':
merge fix for issue bpo-17953
http://hg.python.org/cpython/rev/b60ae4ebf981

Thanks for the patch, Yogesh, but I went with different wording that I came up with on the train the other day while offline. But you did remind me to update Misc/NEWS (which led to a discussion on python-committers), so thanks for that.

You are most welcome Brett. I am just starting to contribute with this doc fix. Didn't think a small detail in it will generate such a long debate in the python-committers mailing list.

Is there a plan in action for 'fixing' Misc/NEWS. I now realize that this change is too small to warrant an entry there, however, as someone new to the list, I simply followed http://docs.python.org/devguide/patch.html. Doing a make patchcheck warns about no changes to Misc/NEWS.

IMHO it would be great if this tool can look at the files changed (or maybe at-least just the directory, or even, simply number of lines modified) and decide if an entry is required in Misc/News. Of-course this is from my perspective. I would like to get your view on this and help about the solution if possible.

I think that is a reasonable suggestion for an enhancement to patchcheck (though I think Brett had a good reason to add a news entry in this particular case). You could open a new issue with that suggestion. On the other hand, patchcheck always warns about other stuff that it can't so easily realize aren't really needed, so it may not be considered a worthwhile enhancement.

That sounds good. I will open up an issue (if not for anything else other than to get more eyes on this issue)

For the record, bpo-12633 has some more discussion on this.