Remove or delay imports to improve startup time

This eliminates the regression in normal startup time vs 2009Q1, though doesn't fix the regression in statup_nosite. There's a lot more headroom (see the comment in site.py), but that will take a slightly more involved fix.

trunk vs patch (default options; perf.py -r)
normal_startup:
Min: 0.435643 -> 0.319087: 36.53% faster
Avg: 0.459272 -> 0.331541: 38.53% faster
Significant (t=51.009351, a=0.95)
Stddev: 0.02377 -> 0.00787: 202.25% smaller

2009Q1 vs patch (default options; perf.py -r)
normal_startup:
Min: 0.381866 -> 0.317074: 20.43% faster
Avg: 0.402331 -> 0.331774: 21.27% faster
Significant (t=20.677263, a=0.95)
Stddev: 0.02758 -> 0.02009: 37.30% smaller

trunk vs patch (-O2 -j always; n=5)
normal_startup:
Min: 15.735776 -> 11.112612: 41.60% faster
Avg: 15.795877 -> 11.144470: 41.74% faster
Significant (t=151.105242, a=0.95)
Stddev: 0.05690 -> 0.03873: 46.93% smaller

[Collin Winter, 2009-06-25 17:14:14]

PTAL

[Jeffrey Yasskin, 2009-06-25 17:59:54]

LGTM. You should probably try to push this upstream.

http://codereview.appspot.com/87063/diff/1/7
File Lib/distutils/sysconfig.py (right):

http://codereview.appspot.com/87063/diff/1/7#newcode95
Line 95: from distutils.errors import DistutilsPlatformError
Comment the lazy imports in this file too?

http://codereview.appspot.com/87063/diff/1/4
File Lib/os.py (right):

http://codereview.appspot.com/87063/diff/1/4#newcode410
Line 410: # But we store them as upper case
What in the world does this comment mean? (No need to fix it.)

http://codereview.appspot.com/87063/diff/1/4#newcode412
Line 412: class _Environ(UserDict.IterableUserDict):
Why not inherit from dict here too? Just because you can't test it?

[Collin Winter, 2009-06-25 18:03:41]

http://codereview.appspot.com/87063/diff/1/7
File Lib/distutils/sysconfig.py (right):

http://codereview.appspot.com/87063/diff/1/7#newcode95
Line 95: from distutils.errors import DistutilsPlatformError
On 2009/06/25 17:59:54, Jeffrey Yasskin wrote:
> Comment the lazy imports in this file too?

Done.

http://codereview.appspot.com/87063/diff/1/4
File Lib/os.py (right):

http://codereview.appspot.com/87063/diff/1/4#newcode410
Line 410: # But we store them as upper case
On 2009/06/25 17:59:54, Jeffrey Yasskin wrote:
> What in the world does this comment mean? (No need to fix it.)

Two separate attempts to say "normalize the case for case-insensitive file
systems", I think.

http://codereview.appspot.com/87063/diff/1/4#newcode412
Line 412: class _Environ(UserDict.IterableUserDict):
On 2009/06/25 17:59:54, Jeffrey Yasskin wrote:
> Why not inherit from dict here too? Just because you can't test it?

Because I can't test it, yes.

[James Abbatiello, 2009-06-25 19:06:50]

http://codereview.appspot.com/87063/diff/1/4
File Lib/os.py (right):

http://codereview.appspot.com/87063/diff/1/4#newcode465
Line 465: class _Environ(dict):
Does this cause problems as mentioned in
http://bugs.python.org/issue2144 or
http://bugs.python.org/issue1367711 ?

http://codereview.appspot.com/87063/diff/1/4#newcode496
Line 496: del self[key]
This calls unsetenv twice for each key