a persistence dance on the buckets/sets it handles directly by itself.
If so, a similar but different change is also needed on the Zope3 branch.

SetIteration structs.

Added lots of comments as a result.  Redid parts of multiunion() because
it was clearly leaking references tucked away inside its SetIteration
struct.  Added XXX comments in another place I'm pretty sure is leaking
similar references in error cases (but am too tired to fix that now, if
so).

this now copies the keys into the work area in one gulp via memcpy,
instead of iterating over them one at a time.  Yields a nice speedup when
it applies (and it usually should apply!).

from a sequence of ints in reverse-sorted order (a quadratic-time
proposition).  That doesn't test anything interesting in context, though.
So fiddled it to do a larger input, but it runs much faster now.

to squash delicate code duplication and for speed.

+ This is optimized in several subtle ways over the current method.  This
  is documented in Maintainer.txt, along with a correctness proof.

+ I'll replace all the BTree searches with this soon.  For now I just
  changed _BTree_get().  I *suspect* this also fixes a subtle bug
  introduced by the recent "don't ignore comparison error" patch:  if a
  comparison did trigger an exception, the function just returned without
  doing the PER_ALLOW_DEACTIVATION + PER_ACCESSED dance.  _BTree_get()
  does that dance again now.

follows from that.

neglected to change the name in new comments barely an hour old.

"value" to "child".  All the rest is mechanical name-changing.  Next
step is to change the decl of child from PyObject* to Sized*.

the common initial fields (pyobject hair, persistence hair, 'size',
and 'len').  I think Jim wants the data decls split into a separate
file, but I'm unclear on why so am not doing that now.

Renamed stuff for consistency.

Added some comments.  More BTree comments will end up in Maintainer.txt.

+ Added interface docs.
+ Gave it a new "newsize" argument, for when the caller can gain speed
  by knowing in advance how many additional elements are needed.  This
  isn't used yet, but will be by multiunion and planned "lopsided merge"
  optimizations (the intent is to grow the bucket big enough in one
  shot, then memcpy a large slice into the bucket).
+ When doubling the size, now sets a MemoryError exception on int
  overflow.

vanished).

long overdue merge of toby-cmp-error-branch. btrees no longer ignore exceptions raised by the key comparison function

new MULTI_INT_UNION.

dumps. Copied source from 1.26.

initial fix causes core dumps. We'll reenable this test when the bug
gets fixed (again).

Also:
    - call shuffle correctly.
    - simplify test_suite()

input sets quickly, and will become much faster again soon (see TODO
in C function multiunion_m).

keys at the beginning of the BTree were deleted.

packtime > revid1's timestamp, in a way that's immune to differences
in platform clock resolution.  Solution given by Tim Peters.

the Standby storage version of the test has a hope of succeeding.
That storage ignores non-sensically out-of-sequence tids so as written
this test wouldn't have passed until after 3220-07-29 10:18:09.882353
<methuselah wink>.

We'll use a tid that we know is earlier than now.

storage needs to be more configurable).

blather log, and optionally replays the transactions into another
storage, timing the differences between the original and the new
storage spead.

objects), read_index would fail when trying to get the max of the
nonexistent oids in the transaction.