new thread had started. At least on my MacBook Pro, that wound up sleeping for
a full 10ms (probably 1 jiffy). By using an Event instead, we can be absolutely
certain that the thread has started, and return more quickly (217us).

Before:
$  ./python.exe -m timeit -s 'from threading import Thread'  't = Thread(); t.start(); t.join()'
100 loops, best of 3: 10.3 msec per loop
$  ./python.exe -m timeit -s 'from threading import Thread; t = Thread()'  't.isAlive()'
1000000 loops, best of 3: 0.47 usec per loop

After:
$  ./python.exe -m timeit -s 'from threading import Thread'  't = Thread(); t.start(); t.join()'
1000 loops, best of 3: 217 usec per loop
$  ./python.exe -m timeit -s 'from threading import Thread; t = Thread()'  't.isAlive()'
1000000 loops, best of 3: 0.86 usec per loop

To be fair, the 10ms isn't CPU time, and other threads including the spawned
one get to run during it. There are also some slightly more complicated ways to
get back the .4us in isAlive() if we want.

sure all tests run even if some fail.

which forbids constructing types that have it set. The effect is to speed

  ./python.exe -m timeit -s 'import abc' -s 'class Foo(object): __metaclass__ = abc.ABCMeta' 'Foo()'

up from 2.5us to 0.201us. This fixes issue 1762.

My tests don't show the promised speed up of 10%. The code is as fast as the old code for simple cases and slightly faster for complex cases with several of args and kwargs. But the patch simplifies the code, too.

like waiting for socket timeouts in test_smtplib :-).

(This may fail on some slow platforms, but we can fix those cases which
should be relatively isolated and easier to find now.)
Move two test cases that didn't require a server to be started
to a separate TestCase.  These tests were taking 3 seconds which
is what the timeout was set to.

Speed this test up by avoiding a sleep and using the event.

failure in the shelve module.  Thanks Thomas Herve.

Thanks Thomas Herve

Thomas Herve explained to me that PyCrypto depends on the constants. I'm adding the aliases because C code for Python 2.x should compile under 2.6 as well. The aliases aren't available in Python 3.x though.

stringobject, unicodeobject and the two formatters are rebuild whenever a header files changes

run the test simultaneously.  The simplest thing I found that worked
on both Windows and Unix was to use the PID.  It's unique so should be
sufficient.  This should prevent many of the spurious failures of
the automated tests since they run as different users.

Also cleanup the directory consistenly in the tearDown methods.

It would be nice if someone ensured that the directories are always
created with a consistent name.

were two module_methods and the one used depended on the order the
modules were loaded.  By making the test module_methods static,
it is not exported and the correct version is picked up.

Thanks Gabriel Genellina.

Suggested by Raymond Hettinger.