This makes the str8 constructor accept the same kinds of types as the
bytes constructor. I had to fix instances of str8("abc") to str8(b"abc")
to make tests pass again. The only remaining failure should be test_str
-- the string test suite must be thoroughly redesigned to fit all three
string-like types.

The changes to str8 look fine.

Now it's time to find out how badly tests break since Georg's patch was
generated.  =)

Attached is an updated patch.  I didn't touch Georg's code short of
tossing in some comments on whether we wanted to sneak around some APIs
for performance or not.  Anything I added was to make all but one test pass.

And that one test is test_str.  There is now this odd mixture of
test_unicode, test_str, and test_bytes.  But that doesn't really fit
anymore since unicode is now str, so that file should probably get
renamed.  And str8 is becoming more like bytes.  That means that
string_tests is really just tests for what is now str and we need a
bytes_tests thing for str8 and bytes.

Assigning back to Guido since he seems to be assigned everything that is
holding up 3.0a2.  =)  If I am lucky I will be able to help with this
testing reorg, but that is iffy.

I'll perhaps find some time today to try and make the tests useful and
consistent again.

Okay, somebody seems to have started this, so I won't interfere for now.

Georg, can you check this in even with test_str failing?

And who is working on that?  If I don't get a name I'll tackle it myself.

Okay, committed in r58640.

Gregory wrote this in test_bytes:
# I've started porting some of those into buffer_tests.py, we should port
# the rest that make sense (the code can be cleaned up to use modern
# unittest methods at the same time).