between calls from the storage and from the out-of-band
invalidation message handler.

on start up, with the default being to wait for the server.

The intent was that it should be possible to bring a client up even
if it can't talk to the server, however, this is a little bit risky if
we need to start them both.  Eventually, we need to be the start-up
code be smart enough to wait if we can't serve data requests during
startup, but to proceed if we can.

a memory leak in mounted databases.

This was to fix a bug caused by an interaction with ZEO that caused a
process to be unable to start if the ZEO client cache contained stale
data.

getting the local and global transaction locks. This has a disastrous
effect, causing updates to be missed locally and causing the
transaction id to be lost.

raise an exception. It should catch and serialize exceptions raised by
the underlying storage, however, unpicklable exceptions could cause it
to accidentally raise an exception. We now pickle exceptions more
carefully, catching and handling pickling errors.

  - Close sockets on process shutdown. This should avoid "address
    already in use" messages on restarts.

  - Log shutdown.

Fixed bugs in handling -u option:

  - The -u option wasn't registered with getopt

  - User lookup failures weren't logged correctly.

Send storage size info to all clients at end of pack and
commit.

understandable.

Sped up startup a tiny bit by not doing synchronous get_info call on
startup. Instead, server will send info asynchrously on startup.

to avoid a security problem arising from the way that Zope
changes the effective user id. The first cache file gets created while
Zope is root, so it can't be reopened as nobody.

mechanism is messed up. In this case, we just close the connection,
which will ultimately cause us to reconnect.

This is important since storea is an asynchronous routine.

- Return error is we get a begin request from a connection that
  already has an active transaction.

- Log connections and disconnection logging. Someone (Sam) decided to
  comment this out of the asyncore code.

- Include connection id in debugging messages.

the connection to the storage server is disconnected.

as root on Unix.

Added a signal handler to (gracefully) close storages when killed.
This causes FileStorages to get their indexes written.

and notification propagation, but i got carried away and tried to fill
in some pieces otherwise slightly mysterious to me...

storing serial numbers in the temporary file holding local transaction
data.

Also added error check in temporary file management to detect
unexpected end-of-file conditions.

the trunk. :(

The bug caused pack to missadjust backward file pointers
in some cases involving versions and undo.

tempfiles do *not* perpetually accumulate.  (TemporaryFile reliably is
removed when the program is done with it, under both windows and
unix.)

ClientCache.__init__(): Set up the first TemporaryFile, and set both
self._p file names to None, to signify use of temp files.

ClientCache.checkSize(): Switch over to other persistent cache file if
self._p has names, or a new TemporaryFile if they're None.

(I included some comments to the code i touched, to ease the passage
of anyone else who winds up on that path...)