The following source file:
"""
import _winreg
_winreg.OpenKey(_winreg.HKEY_LOCAL_MACHINE, "foo")
"""

results in the following "patch":

-import _winreg
-_winreg.OpenKey(_winreg.HKEY_LOCAL_MACHINE, "foo")
+import winreg
+winreg.OpenKey(_winreg.HKEY_LOCAL_MACHINE, "foo")

Note the first param has not been converted correctly.  This is probably
due to it following the paren, as using the same symbol in a local
variable works correctly.

I'll look in to it. Just give me a couple of days.

The problem is that fix_imports doesn't look at matches that are nested
within matches. So the _winreg.OpenKey part gets fixed, but the
_winreg.HKEY_LOCAL_MACHINE does not because it is nested within the
other node. I didn't make fix_imports so I don't know what the intention
was for not matching nested matches. When I removed that restriction,
the fixer works as expected. It also does not cause any of 2to3's tests
to fail nor does it have a noticeable impact on 2to3's runtime, so I'm
tempted to say that this is the fix to make, but I don't want to commit
to it until I've heard from it's author about it. I added Collin to the
Nosy List, so hopefully he can comment on the reasoning behind the
restriction.

Here is a patch with a new test case. I don't know if we should apply it
until we've heard from Collin though. The old version of fix_imports
fails the new test, but this version, which allows matching of nodes
nested within matches, passes the new test I added for nesting, as well
as all of the other 2to3 tests.

I do have performance problems if I remove the match() override, but I
found that special casing usage replacements fixes the bug while keeping
speed.

Done in r67390. Nick, thanks for the debugging and the patch!