A large number of changes so that cache tracing isn't
100% broken -- I can get a trace file now, and stats.py
doesn't blow up anymore.  Don't yet know whether it's
reporting sensible things, but it's not obviously insane.
simul.py doesn't know about the current cache implementation
either yet.

ClientCache:  made it a new-style class.

FileCache:  Removed the reuse= argument.  IIRC, it made some sense
when different cache schemes were first being tried, but complicated
the logic now to no good end.

Nuked the maddening log warnings about reuse=True.  They were never
helpful and were often confusing.

Added info-level log messages to record the path of the client cache
file, whether it's persistent or temporary, and if it's persistent
whether we're creating or reusing it.

Minor changes to comments.

Massive rewrite of cache.txt, to match current reality.

Collector 1843:  IISet.keys: bad exception handling.

Bucket_rangeSearch():  De-obfuscated the calls to
Bucket_findRangeEnd(), so that they stop ignoring
the latter's error returns (a mind-bending combination
of embedded assignment nested in an UNLESS macro,
seemingly copy+paste'd so that the error occurred twice).

Port from ZODB 3.2.

Collector #1846:  If an uncommitted transaction was found, fsrecover.py
fell into an infinite loop.  Fixed that, and added a new test
(testUncommittedAtEnd) to ensure this stays fixed.

Subtransactions are deprecated in 3.5, and will go away
in 3.7.

Convert internal uses of subtxns to use savepoints instead.

I suspect BTrees/convert.py should be removed instead.

Revert extra copying in Connection._rollback().

Since TmpStore.reset() had to change too to copy the index anyway,
it no longer hurts that _rollback() clears the index (as a side
effect of self._cache.invalidate(src.index)).

Merge tim-savepoint branch.

Allow rollback to a given savepoint multiple times.

transaction/_transaction.py
    _remove_and_invalidate_after():  Don't remove the
    argument savepoint from the savepoint stack anymore.

    abort():  abort(True) relied on the subtxn savepoint
    being invalidated as a side effect of rolling back
    to it.  Invalidate it explicitly now.

    Savepoint.rollback():  Don't invalidate self.

ZODB/Connection.py
    Connection._rollback():  Pass a temp container to
    _cache.invalidate(), because that method clears the
    container as a side effect.  Clearing the index
    in a savepoint was a disaster if the savepoint was
    used again.

    TmpStore.reset():  Make a copy of the argument index.
    An alternative would be to make all callers make
    copies, but the only caller now thinks of its `state`
    argument as being opaque.

transaction/savepoint.txt
ZODB/tests/{testSubTransaction.py, testConnectionSavepoint.txt}
    Test the new semantics, and beef up the tests.

transaction/tests/savepointsample.py
    _rollback_savepoint():  Needed to copy the data.

English repairs.

Simplify.

Primarily, _tindex should have been a dict all along, pack()
worked too hard at the end, and loadEx() is a minor variant
of load().

load() and loadEx():  eliminate gratuitous differences.

Ideally, load() should call loadEx() instead, but we really
don't want "an extra" Python-level call here (heavily used).

loadEx():  deleted pointless call of self._read_txn_header().

ServerStub.loadEx() comments:  these were obviously wrong in
several ways, but I don't know the full truth.  Better to
say so up front than to leave them clearly wrong, though.

Collector 1831.

The BTree minKey() and maxKey() methods gave a misleading message if no key
satisfying the constraints existed in a non-empty tree.

Collector #1830.

In some error cases when reading a FileStorage index, the
code referenced an undefined global.

Get rid of all code defining or referencing t32.

Such code never made sense; it was all of the form

    t32 = 1L << 32
    if something < 0:
        something = t32 - something

which is a way to change negative integers into
gigantic positive integers greater than 4 billion, and
never made sense in context.  Possibly

        something = t32 + something

was intended, but that never made sense in context either.

In any case, `something` is invariably obtained from
struct.unpack using the "H" format code, and that never
returns a negative integer to begin with.

read_index():  Mechanical reformat for easier reading.

Collector 1829.

Clarified that the ``minKey()`` and ``maxKey()`` methods raise an exception
if no key exists satsifying the constraints.

Also improved the English in other interface docstrings.

Collector 1822.

Make undo{Log,Info} arguments act like Python slice indices when
both are non-negative.  The code used to do that before ZODB 3.4a9,
but changed to match ZODB's UML documentation.  Alas, some
(untested) code in Zope relied on the actual behavior (see the
collector report).  Changed code, docs, and tests to bless the
old behavior in these cases.

DemoStorage.UndoLog:  this was wrong in several ways.  I'm still
unsure about why it skips "packed" transactions.  That doesn't seem
right, but I don't have time to wonder about that now.

_triggerbase:  Make new-style class, for better debugability.

Windows trigger.__init__:  Simplify more.  The trigger can connect
after the other end has done  bind() and listen(); no need to wait
for an accept() too, and so no need to fiddle with blocking/
non-blocking or "expected" socket.error's either.

Largely rewritten.

remove_loop_callback():  There were two definitions of this function.
Deleted one, changed the other to use enumerate().

poll():  This was trying to repair a bug in Python 2.2's asyncore.poll().
2.2 is no longer supported, so removed this poll().

_start_loop(), _stop_loop():  Removed; functionality folded into
reworked loop().

loop():  The signature of asyncore.loop() changed in 2.4, but this
still had "the old" signature.  Instead of sniffing the signature
of the Python in use, this does a saner thing:  the original
asyncore.loop is captured, and called from the body of this loop().
That way, (A) we don't care what asyncore.loop's signature is; and,
(B) we still call the original Python code, so debugging prints and
breakpoints (etc) stuffed into Python's asyncore no longer "vanish
by magic" when ZEO is used.

Massive refactoring, to move the bulk of the trigger code into
an OS-indepedent base class.

__repr__:  Use the postive_id function to embed the machine address.
Addresses with the high bit set trigger warnings before Python 2.4,
and come out as negative numbers in 2.4+.

Windows trigger.__init__:  Don't make 50 guesses at a port number
to use, let Windows pick an available port for us.  Also documented
the baffling single-thread socket setup dance, which took an hour to
reverse-engineer (in large part because it used a bare "except" w/ no
clue as to why).

reachable from multiple connections.  Previously, we thought that we
could limit the time that an object was new to a single savepoint, but
that didn't work because savepoints of different connections are too
independent.   Now an object is considered new for the full extent of
the transaction in which it was created.

Made it possible to use connection add methods to explicitly control
the database an object is added too.

TmpStore needs to delegate loadBefore().

Remove needless "cleanup" code -- Transaction does not have
a __del__ method (but probably did at some earlier time).

1. Construction of Connections is an implementation detail.
   ZODB client code should never construct connections directly.

2. __init__ is not a method of IConnection.  It might be a method of
   some sort of connection factory api, if we had one.

connections.   Connections now get their database and related data
when they are constructed.  They no longer throw away this information
when they are closed.  (The old design that threw away the database
and related data did so to avoid memory leaks in Python 1.)

Fixed a multi-database connection-management bug. (The fix
necessitated the simplification.)

by virtue of being reachable from objects in databases, there is
a danger of ambiguity when an object is reachable from multiple
databases.  This ambiguity can lead to very subtle bugs with the
symtom that objects are in unexpected databases.

Introduced a check to refuse to commit when the ambiguity exists.

(This required changing the connection '_creating' attribute to a
dictionary.)

Secondary connections always need to be made from an existing
connection.

Added check to make sure we don't get references across incompatible
connections.

pass optional open arguments to secondary databases.

savepoint.