admin address of the subprocess.  Better to gulp 'em now than sip 'em
later.

shutdown_zeo_server(): Same thing here, but notice that on Windows,
the recv() can raise an ECONNRESET, which I believe we can just
ignore.  (Will test this with Tim shortly.)

Instead use self.new_dest() to get a new destination storage.

XXX It would still be useful to check for cross-storage-implementation
restores.  IOW, fs->bdb, bdb->fs (others?).

the warnings until the other fixes are ported to ZODB4.

Added an XXX to memind me.

function.  This was a nasty decoy (am I using that term correctly?
:).

Because we're worried about backward compatibility, for now we'll
rebind asyncore.loop() to a function that prints a warning and then
calls loop().  Eventually we'll take this away so you have to call
ThreadedAsync.loop() explicitly.

through the sneakily rebound asyncore.loop().

rename UnixTests to FileStorageTests, and add a simple BDBTests class
to use Berkeley Full storage.

Rewrite to use the new forker module api.

WindowsTests: Removed.

rename UnixConnectionTests to FileStorageConnectionTests, and add a
simple BDBConnectionTests class to use Berkeley Full storage.

WindowsConnectionTests: Removed.

DummyDB.invalidate(): Add **kws since some of the tests try to use
keyword arguments.

Removed redundant keyword arguments from various openClientStorage()
calls.

makes little sense to call the server script `winserver.py'.  That's
gone now, in favor of this zeoserver.py script.  Changes include:

main(): Accept a -C option pointing to a temp file containing a
ZConfig storage section.  This reads the file and creates the so
described storage, then removes the temp config file.  It creates an
administrative server which opens an `admin' port which is used to
os._exit() this process, with the following protocol.  The first time
the admin port is connected to, an ack character is sent from
zeoserver.  This is necessary to coordinate the tests so that they are
assured the server is running and responding before it tries to do
more.  Otherwise, some tests were vulnerable to timing bugs where the
shutdown connect happened before the server was ready to accept them.

The second connect exits the server process just like before.  Copious
log entries are written.

This script also cleans up after the storages are closed, by calling
the storage's .cleanup() method if it exists (FileStorage and the
Berkeley storages both implement this method).

and Unix.  We actually use the Windows code as the basis for the new
approach so there's some hope that this will actually work on Windows
<wink>.  Tim will verify once the checkins are finished.

start_zeo_server(): Is now the blessed way of starting a ZEO server,
and it primarily takes a ZConfig storage section string as its
argument.  This it writes to a temp file and passes -C to the
zeoserver.py script, which parses that temp file and creates the
underlying storage.  zeoserver.py is responsible for deleting the temp
file.

zeoserver opens an `admin' port which is used to quit the server, but
the protocol has been expanded slightly.  The first time the admin
port is connected to, an ack character is sent from zeoserver.  This
is necessary to coordinate the tests so that they are assured the
server is running and responding before it tries to do more.
Otherwise, some tests were vulnerable to timing bugs where the
shutdown connect happened before the server was ready to accept them.

The second connect exits the server process just like before.

start_zeo_server() returns a 3-tuple.  The first element is the
(host,port) tuple of the ZEO server, the second is the (host,port)
tuple of the admin socket, and the third is the pid of the spawned
child process.

shutdown_zeo_server(): The blessed way to shut down a zeoserver.py
process.  Takes an admin (host,port) tuple, connects to that and
writes a log entry.

The get_class() function will not find a class for a ZClass, but it
shouldn't fail.  Fix this by making the class an output paramter got
get_class() and returning 1/0 for success / failure.  A ZClass should
succeed trivially because it is not an extension class.

asyncore.loop() which really isn't.

tests, including read_only tests.

create the specific storage instances.  This lets us easily generalize
the tests (the Unix ones at least ;) to use the Berkeley Full storage.

module function cleanup() which was added to FileStorage and
BerkeleyBase.

UnixTests.getStorage(): Return a ZConfig string instead of a tuple of
a class name and an argument list.

BDBTests: Simple extension to UnixTests to test using Berkeley
storages with ZEO.

WindowsTest: Derive from UnixTests and remove getStorageInfo() and
delStorage().

Add BDBTests if we can import Full.

create the specific storage instances.  This lets us easily generalize
the tests (the Unix ones at least ;) to use the Berkeley Full storage.

Tim and I will check Windows compatibility soon.

i.e. reason first, then oid.  This is because there's always a reason
but sometimes there isn't an oid (so we can default arg the oid).

i.e. reason first, then oid.  This is because there's always a reason
but sometimes there isn't an oid (so we can default arg the oid).

cleanup(): Add both a method to FileStorage and a module global
function to clean up all the files that a FileStorage creates.
Sometimes it's handy to do a cleanup if you have the instance, other
times you only have the Data.fs file name.

PersistentMapping sitting off the root object.  The zrs heartbeat
object will go in this same location (under a different name of
course).

FileStorage._txn_undo()

instead now.

The _txn_find() must not stop at the pack boundary when it is called
by restore().  It was originally written for _txn_undo() which isn't
supposed to undo to a transaction across a pack.  But it should be
legal to restore() a transaction with a reference to a data record in
a transaction that was packed.

Fix by adding stop_at_pack flag to _txn_find() and add tests of this
behavior for FileStorage.

checkTransactionExtensionFromIterator(): New test which BDBFullStorage
was failing because it wasn't unpickling the transaction extension
data.

Oops.  The code for BTrees-based indexing was done on a branch, which
was later merged into another branch, which was later merged to the
trunk.  Somewhere along the way the change that actually enabled the
new index was lost.

Almost entirely concerned with adding ZPL blocks.

The timeout is necessary because in some cases a non-blocking connect
while neither succeed nor fail.  It is unacceptable to simply get
stuck forever in this code, because it is run at least once inside the
ClientStorage constructor.

Set the first timeout to a fairly low number -- 5 seconds -- so that a
ClientStorage(wait=0) call does not take a inordinate length of time.
Then set the default to a longer number -- 75 seconds, which is
typical timeout for a blocking connect() call.

This change fixes the hang in testZEOStorage from testStorageConfig.

XXX Need to get Guido to review the changes.