Fix SF bug #516768 reported by Dave Wallace.

Replace use of PyMem_DEL() with PyObject_Del() on object dealloc
functions.  The use of PyMem_DEL() is incorrect for object
deallocation, because it only ever calls the low-level free().  If a
custom allocator like pymalloc is used, it needs to be called to free
the memory.

Also replace bare malloc() and realloc() with PyMem_Malloc() and
PyMem_Realloc().  I think this isn't a strict bug fix, but a would be
nice.  It guarantees that BTrees objects get their memory from the
same allocator as the Python core.

   The previous, insane version was passing None or a thread id to "%.03f".

Add some whitespace around get_transaction() implementations.

The code was intended to be added to the Zope3 branch, not modified on
the trunk.

of the StandaloneZODB-1_0-branch into the trunk.  The
StandaloneZODB-1_0-branch is now dead!

trunk.  I'm *almost* done with the merge; the only file not yet merged
is FileStorage.py.

code that is needed for it yet.

Summary of changes:

- Add function gc_all_items() to call gc_item() on everything in
  dict.  This factors out common code in all the gc methods.

- Add check_size() function that factors out code to check cache size
  before starting a gc.

- Simplify argument processing in cc_invalidate(), avoiding a second
  call to PyArg_ParseTuple() when argument is a sequence instead of a
  dictionary.

- Chance code to reflect belief that NULL return from PyDict_GetItem()
  does not set exception, except in cases that don't apply for the
  pickle cache like comparing a string with high-order bit set to a
  Unicode object.

- Use if (!expr) instead of UNLESS().

- Remove all statements from conditionals.

- Reformat to 4 spaces and Python-style brace placement.

- Put whitespace around operators and after commas.

Refactor main() to report all errors and exit with non-zero status.

Remove useless -d option.

the trunk.  Sigh, if I'd known about that, I wouldn't have written
another test suite for TimeStamp. :-(

aborted.

integers, due to MS's slothful 18.2Hz updating of time.time().  The test
implicitly assumed in one spot that "the midpoint" between two revids was
strictly larger than the smaller one, but that wasn't true on Windows.
Adjusted the computation (as suggested by Guido) so that "strictly larger"
holds in the end even when the inputs are adjacent integers.

the code

    try:
        root = self._getRoot()
        ...
    except:
        self._closeDB(root)
	self._delDB()
	raise

which fails with an UnboundLocalError if the first line in the try
clause fails.  Fixed this by setting 'root = None' before each such
try clause, and adding a None check to _closeDB().  Also removed a
useless 'root = None' from the body of _closeDB().

Barry: I saw this in the saz 1.0c1 release.  Is it worth backporting?
Probably not given that we're not going to do a merge pre-1.0.

Put the tpc_begin() -> tpc_finish() in a try/except.  If an error
occurs, abort the current transaction and re-raise the error.

It appears that some storages (well, only Standby Storage) has trouble
shutting down if a transaction is in progress.

Replace three small integers with symbolic constants.
    TRANS_HDR_LEN = 23
    DATA_HDR_LEN = 42
    DATA_VERSION_HDR_LEN = 58
The constants make it easier to remember what kind of header is being
read and/or what kind of offset is being computed.

Refactor commitVersion():
    - Keep locking & arg checking in commitVersion(), but move actual
      commit logic to _commitVersion().  This removes a level of
      indentation.

    - Remove local vars that alias attributes of other vars

Add a few more POSKeyError exceptions.