A number of tests are designed to be skipped on 64bits, but they don't
detect 64bit windows builds as 64bits.  Also, test_bytes.test_repeat()
assumes sys.maxint bytes can't be allocated, which isn't necessarily
true on Win64.

I'm attaching a patch that arranges to skip these tests.

Mark, that check looks fragile:

+IS32BIT = sys.maxint == 0x7fffffff and "64 bit" not in sys.version

Why don't you check for sys.maxsize instead? By construction,
sys.maxsize should give you the pointer width (minus the sign bit ;-)).

I forgot about sys.maxsize - that makes things much cleaner, and even
means I don't need to skip the check for insane amounts of memory.

If you think this is OK, please also specifically say if you approve for
me to check it into trunk for 2.6.  Thanks.

Mark, your patch is ok to me. You can check it in.

Thanks.  Checked into the trunk in r65986