Metadata-Version: 1.0
Name: zc.twist
Version: 1.3.1
Summary: Mixing Twisted and ZODB

Home-page: http://pypi.python.org/pypi/zc.twist
Author: Zope Project
Author-email: zope-dev@zope.org
License: ZPL
Description: ===================================================
        Twist: Talking to the ZODB in Twisted Reactor Calls
        ===================================================
        
        The twist package contains a few functions and classes, but primarily a
        helper for having a deferred call on a callable persistent object, or on
        a method on a persistent object.  This lets you have a Twisted reactor
        call or a Twisted deferred callback affect the ZODB.  Everything can be
        done within the main thread, so it can be full-bore Twisted usage,
        without threads.  There are a few important "gotchas": see the Gotchas_
        section below for details.
        
        The main API is `Partial`.  You can pass it a callable persistent object,
        a method of a persistent object, or a normal non-persistent callable,
        and any arguments or keyword arguments of the same sort.  DO NOT
        use non-persistent data structures (such as lists) of persistent objects
        with a database connection as arguments.  This is your responsibility.
        
        If nothing is persistent, the partial will not bother to get a connection,
        and will behave normally.
        
            >>> from zc.twist import Partial
            >>> def demo():
            ...     return 42
            ...
            >>> Partial(demo)()
            42
        
        Now let's imagine a demo object that is persistent and part of a
        database connection.  It has a `count` attribute that starts at 0, a
        `__call__` method that increments count by an `amount` that defaults to
        1, and an `decrement` method that reduces count by an `amount` that
        defaults to 1 [#set_up]_.  Everything returns the current value of count.
        
            >>> demo.count
            0
            >>> demo()
            1
            >>> demo(2)
            3
            >>> demo.decrement()
            2
            >>> demo.decrement(2)
            0
            >>> import transaction
            >>> transaction.commit()
        
        Now we can make some deferred calls with these examples.  We will use
        `transaction.begin()` to sync our connection with what happened in the
        deferred call.  Note that we need to have some adapters set up for this
        to work.  The twist module includes implementations of them that we
        will also assume have been installed [#adapters]_.
        
            >>> call = Partial(demo)
            >>> demo.count # hasn't been called yet
            0
            >>> deferred = call()
            >>> demo.count # we haven't synced yet
            0
            >>> t = transaction.begin() # sync the connection
            >>> demo.count # ah-ha!
            1
        
        We can use the deferred returned from the call to do something with the
        return value.  In this case, the deferred is already completed, so
        adding a callback gets instant execution.
        
            >>> def show_value(res):
            ...     print res
            ...
            >>> ignore = deferred.addCallback(show_value)
            1
        
        We can also pass the method.
        
            >>> call = Partial(demo.decrement)
            >>> deferred = call()
            >>> demo.count
            1
            >>> t = transaction.begin()
            >>> demo.count
            0
        
        This also works for slot methods.
        
            >>> import BTrees
            >>> tree = root['tree'] = BTrees.family32.OO.BTree()
            >>> transaction.commit()
            >>> call = Partial(tree.__setitem__, 'foo', 'bar')
            >>> deferred = call()
            >>> len(tree)
            0
            >>> t = transaction.begin()
            >>> tree['foo']
            'bar'
        
        Arguments are passed through.
        
            >>> call = Partial(demo)
            >>> deferred = call(2)
            >>> t = transaction.begin()
            >>> demo.count
            2
            >>> call = Partial(demo.decrement)
            >>> deferred = call(amount=2)
            >>> t = transaction.begin()
            >>> demo.count
            0
        
        They can also be set during instantiation.
        
            >>> call = Partial(demo, 3)
            >>> deferred = call()
            >>> t = transaction.begin()
            >>> demo.count
            3
            >>> call = Partial(demo.decrement, amount=3)
            >>> deferred = call()
            >>> t = transaction.begin()
            >>> demo.count
            0
        
        Arguments themselves can be persistent objects.  Let's assume a new demo2
        object as well.
        
            >>> demo2.count
            0
            >>> def mass_increment(d1, d2, value=1):
            ...     d1(value)
            ...     d2(value)
            ...
            >>> call = Partial(mass_increment, demo, demo2, value=4)
            >>> deferred = call()
            >>> t = transaction.begin()
            >>> demo.count
            4
            >>> demo2.count
            4
            >>> demo.count = demo2.count = 0 # cleanup
            >>> transaction.commit()
        
        ConflictErrors make it retry.
        
        In order to have a chance to simulate a ConflictError, this time imagine
        we have a runner that can switch execution from the call to our code
        using `pause`, `retry` and `resume` (this is just for tests--remember,
        calls used in non-threaded Twisted should be non-blocking!)
        [#conflict_error_setup]_.
        
            >>> import sys
            >>> demo.count
            0
            >>> call = Partial(demo)
            >>> runner = Runner(call) # it starts paused in the middle of an attempt
            >>> call.attempt_count
            1
            >>> demo.count = 5 # now we will make a conflicting transaction...
            >>> transaction.commit()
            >>> runner.retry()
            >>> call.attempt_count # so it has to retry
            2
            >>> t = transaction.begin()
            >>> demo.count # our value hasn't changed...
            5
            >>> runner.resume() # but now call will be successful on the second attempt
            >>> call.attempt_count
            2
            >>> t = transaction.begin()
            >>> demo.count
            6
        
        By default, after five ConflictError retries, the partial fails, raising the
        last ConflictError.  This is returned to the deferred.  The failure put
        on the deferred will have a sanitized traceback.  Here, imagine we have
        a deferred (named `deferred`) created from such a an event
        [#conflict_error_failure]_.
        
            >>> res = None
            >>> def get_result(r):
            ...     global res
            ...     res = r # we return None to quiet Twisted down on the command line
            ...
            >>> d = deferred.addErrback(get_result)
            >>> print res.getTraceback() # doctest: +ELLIPSIS +NORMALIZE_WHITESPACE
            Traceback (most recent call last):
            ...
            ZODB.POSException.ConflictError: database conflict error...
        
        You can control how many ConflictError (and other transaction error) retries
        should be performed by setting the ``max_transaction_errors`` attribute
        [#max_transaction_errors]_.
        
        ZEO ClientDisconnected errors are always retried, with a backoff that, by
        default begins at 5 seconds and is never greater than 60 seconds
        [#relies_on_twisted_reactor]_ [#use_original_demo]_ [#client_disconnected]_.
        
        Other errors are returned to the deferred, like a transaction error that has
        exceeded its available retries, as sanitized failures.
        
            >>> call = Partial(demo)
            >>> d = call('I do not add well with integers')
            >>> d = d.addErrback(get_result)
            >>> print res.getTraceback() # doctest: +ELLIPSIS +NORMALIZE_WHITESPACE
            Traceback (most recent call last):
            ...
            ...TypeError: unsupported operand type(s) for +=: 'int' and 'str'
        
        The failure is sanitized in that the traceback is gone and the frame values
        are turned in to reprs.  If you pickle the failure then it truncates the
        reprs to a maximum of 20 characters plus "[...]" to indicate the
        truncation [#show_sanitation]_.
        
        The call tries to be a good connection citizen, waiting for a connection
        if the pool is at its maximum size.  This code relies on the twisted
        reactor; we'll use a `time_flies` function, which takes seconds to move
        ahead, to simulate time passing in the reactor.
        
        We use powers of 2 for the floating-points numbers (e.g. 0.125) to avoid
        a floating-point additive accumulation error that happened in the tests
        when values such as 0.1 were used.
        
            >>> db.setPoolSize(1)
            >>> db.getPoolSize()
            1
            >>> demo.count = 0
            >>> transaction.commit()
            >>> call = Partial(demo)
            >>> res = None
            >>> deferred = call()
            >>> d = deferred.addCallback(get_result)
            >>> call.attempt_count
            0
            >>> time_flies(.125) >= 1 # returns number of connection attempts
            True
            >>> call.attempt_count
            0
            >>> res # None
            >>> db.setPoolSize(2)
            >>> db.getPoolSize()
            2
            >>> time_flies(.25) >= 1
            True
            >>> call.attempt_count > 0
            True
            >>> res
            1
            >>> t = transaction.begin()
            >>> demo.count
            1
        
        If it takes more than a second or two, it will eventually just decide to grab
        one.  This behavior may change.
        
            >>> db.setPoolSize(1)
            >>> db.getPoolSize()
            1
            >>> call = Partial(demo)
            >>> res = None
            >>> deferred = call()
            >>> d = deferred.addCallback(get_result)
            >>> call.attempt_count
            0
            >>> time_flies(.125) >= 1
            True
            >>> call.attempt_count
            0
            >>> res # None
            >>> time_flies(2) >= 2 # for a total of at least 3
            True
            >>> res
            2
            >>> t = transaction.begin()
            >>> demo.count
            2
        
        Without a running reactor, this functionality will not work
        [#teardown_monkeypatch]_.
        
        You can also specify a reactor for the partial using ``setReactor``, if
        you don't want to use the standard one installed by twisted in
        ``twisted.internet.reactor``. [#setReactor]_
        
        Gotchas
        -------
        
        For a certain class of jobs, you won't have to think much about using
        the twist Partial.  For instance, if you are putting a result gathered by
        work done by deferreds into the ZODB, and that's it, everything should be
        pretty simple.  However, unfortunately, you have to think a bit harder for
        other common use cases.
        
        * As already mentioned, do not use arguments that are non-persistent
          collections (or even persistent objects without a connection) that hold
          any persistent objects with connections.
        
        * Using persistent objects with connections but that have not been
          committed to the database will cause problems when used (as callable
          or argument), perhaps intermittently (if a commit happens before the
          partial is called, it will work).  Don't do this.
        
        * Do not return values that are persistent objects tied to a connection.
        
        * If you plan on firing off another reactor call on the basis of your
          work in the callable, realize that the work hasn't really "happened"
          until you commit the transaction.  The partial typically handles commits
          for you, committing if you return any result and aborting if you raise
          an error. But if you want to send off a reactor call on the basis of a
          successful transaction, you'll want to (a) do the work, then (b)
          commit, then (c) send off the reactor call.  If the commit fails,
          you'll get the standard abort and retry.
        
        * If you want to handle your own transactions, do not use the thread
          transaction manager that you get from importing transaction.  This
          will cause intermittent, hard-to-debug, unexpected problems.  Instead,
          adapt any persistent object you get to
          transaction.interfaces.ITransactionManager, and use that manager for
          commits and aborts.
        
        =========
        Footnotes
        =========
        
        .. [#set_up] We'll actually create the state that the text describes here.
        
            >>> import persistent
            >>> class Demo(persistent.Persistent):
            ...     count = 0
            ...     def __call__(self, amount=1):
            ...         self._p_deactivate() # to be able to trigger ClientDisconnected
            ...         self.count += amount
            ...         return self.count
            ...     def decrement(self, amount=1):
            ...         self.count -= amount
            ...         return self.count
            ...
            >>> from ZODB.tests.util import DB
            >>> db = DB()
            >>> conn = db.open()
            >>> root = conn.root()
            >>> demo = root['demo'] = Demo()
            >>> demo2 = root['demo2'] = Demo()
            >>> import transaction
            >>> transaction.commit()
        
        .. [#adapters] You must have two adapter registrations: IConnection to
            ITransactionManager, and IPersistent to IConnection.  We will also
            register IPersistent to ITransactionManager because the adapter is
            designed for it.
        
            >>> from zc.twist import transactionManager, connection
            >>> import zope.component
            >>> zope.component.provideAdapter(transactionManager)
            >>> zope.component.provideAdapter(connection)
            >>> import ZODB.interfaces
            >>> zope.component.provideAdapter(
            ...     transactionManager, adapts=(ZODB.interfaces.IConnection,))
        
            This quickly tests the adapters:
        
            >>> ZODB.interfaces.IConnection(demo) is conn
            True
            >>> import transaction.interfaces
            >>> transaction.interfaces.ITransactionManager(demo) is transaction.manager
            True
            >>> transaction.interfaces.ITransactionManager(conn) is transaction.manager
            True
        
        .. [#conflict_error_setup] We also use this runner in the footnote below.
        
            >>> import threading
            >>> _main = threading.Lock()
            >>> _thread = threading.Lock()
            >>> class AltDemo(persistent.Persistent):
            ...     count = 0
            ...     def __call__(self, amount=1):
            ...         self.count += amount
            ...         assert _main.locked()
            ...         _main.release()
            ...         _thread.acquire()
            ...         return self.count
            ...
            >>> demo = root['altdemo'] = AltDemo()
            >>> transaction.commit()
            >>> class Runner(object):
            ...     def __init__(self, call):
            ...         self.call = call
            ...         self.thread = threading.Thread(target=self.run)
            ...         self.thread.setDaemon(True)
            ...         _thread.acquire()
            ...         _main.acquire()
            ...         self.thread.start()
            ...         _main.acquire()
            ...     def run(self):
            ...         self.running = True
            ...         self.result = self.call()
            ...         assert _main.locked()
            ...         assert _thread.locked()
            ...         _thread.release()
            ...         self.running = False
            ...         _main.release()
            ...     def retry(self):
            ...         assert _thread.locked()
            ...         _thread.release()
            ...         _main.acquire()
            ...     def resume(self):
            ...         while self.running:
            ...             self.retry()
            ...         self.thread.join()
            ...         assert not _thread.locked()
            ...         assert _main.locked()
            ...         _main.release()
        
        .. [#conflict_error_failure] Here we create five consecutive conflict errors,
            which causes the call to give up.
        
            >>> call = Partial(demo)
            >>> runner = Runner(call)
            >>> for i in range(5):
            ...     demo.count = i
            ...     transaction.commit() # creates a write conflict
            ...     runner.retry()
            ...
            >>> demo.count
            4
        
            When we resume without a conflict error, it is too late: the result is a
            ConflictError.  The ConflictError is actually shown in the main text.
        
            >>> runner.resume()
            >>> demo.count
            4
            >>> call.attempt_count
            5
            >>> deferred = runner.result
        
        .. [#max_transaction_errors] As the main text mentions, the
           ``max_transaction_errors`` attribute lets you set how many conflict errors
           should be retried.
        
            >>> call = Partial(demo)
            >>> call.max_transaction_errors
            5
            >>> call.max_transaction_errors = 10
            >>> call.max_transaction_errors
            10
            >>> runner = Runner(call)
            >>> for i in range(10):
            ...     demo.count = i
            ...     transaction.commit()
            ...     runner.retry()
            ...
        
           When we resume without a conflict error, it is too late: the result is a
           ConflictError.
        
            >>> runner.resume()
            >>> demo.count
            9
            >>> call.attempt_count
            10
            >>> deferred = runner.result
        
            >>> res = None
            >>> def get_result(r):
            ...     global res
            ...     res = r # we return None to quiet Twisted down on the command line
            ...
            >>> d = deferred.addErrback(get_result)
            >>> print res.getTraceback() # doctest: +ELLIPSIS +NORMALIZE_WHITESPACE
            Traceback (most recent call last):
            ...
            ZODB.POSException.ConflictError: database conflict error...
        
           Setting ``None`` means to retry conflict errors forever.  For our example,
           we will arbitrarily choose 50 iterations to show "forever".
        
            >>> call = Partial(demo)
            >>> call.max_transaction_errors
            5
            >>> call.max_transaction_errors = None
            >>> print call.max_transaction_errors
            None
            >>> runner = Runner(call)
            >>> for i in range(50):
            ...     demo.count = i
            ...     transaction.commit()
            ...     runner.retry()
            ...
        
           Now when we resume without a conflict error, we get a successful result: it
           never gave up.
        
            >>> runner.resume()
            >>> runner.result # doctest: +ELLIPSIS
            <Deferred at ...  current result: 50>
            >>> _ = transaction.begin() # we need to sync to get changes
            >>> demo.count # notice, 49 + 1
            50
        
        .. [#relies_on_twisted_reactor] We monkeypatch twisted.internet.reactor
            (and revert it in another footnote below).
        
            >>> import twisted.internet.reactor
            >>> oldCallLater = twisted.internet.reactor.callLater
            >>> import bisect
            >>> class FauxReactor(object):
            ...     def __init__(self):
            ...         self.time = 0
            ...         self.calls = []
            ...     def callLater(self, delay, callable, *args, **kw):
            ...         res = (delay + self.time, callable, args, kw)
            ...         bisect.insort(self.calls, res)
            ...         # normally we're supposed to return something but not needed
            ...     def time_flies(self, time):
            ...         end = self.time + time
            ...         ct = 0
            ...         while self.calls and self.calls[0][0] <= end:
            ...             self.time, callable, args, kw = self.calls.pop(0)
            ...             callable(*args, **kw) # normally this would get try...except
            ...             ct += 1
            ...         self.time = end
            ...         return ct
            ...
            >>> faux = FauxReactor()
            >>> twisted.internet.reactor.callLater = faux.callLater
            >>> time_flies = faux.time_flies
        
        .. [#use_original_demo] The second demo has too much thread code in it:
            we'll use the old demo for the rest of the discussion.
        
            >>> demo = root['demo']
        
        .. [#client_disconnected] As the main text describes,
           ZEO.Exceptions.ClientDisconnected errors will always be retried, but with a
           backoff.
        
           First we'll mimic a disconnected ZEO at the start of a transaction.
        
            >>> from ZEO.Exceptions import ClientDisconnected
            >>> raise_error = [1]
            >>> storage_class = db._storage.__class__
            >>> original_load = storage_class.load
            >>> def load(self, oid, version):
            ...     if raise_error:
            ...         raise_error.pop()
            ...         raise ClientDisconnected()
            ...     else:
            ...         return original_load(self, oid, version)
            ...
            >>> _ = transaction.begin()
            >>> demo.count
            0
            >>> call = Partial(demo)
            >>> storage_class.load = load
        
           We rely on a reactor to implement delayed calls.  We have a fake reactor
           called ``faux`` for these examples.  It has a list of pending calls, and
           we can call ``time_flies`` to make time appear to pass.
        
            >>> len(faux.calls)
            0
        
           When we first call the partial, it will fail, and reschedule for later.
        
            >>> len(faux.calls)
            0
            >>> deferred = call()
            >>> deferred.called
            0
            >>> len(faux.calls)
            1
        
           The rescheduling is initially for five seconds later, by default.  In this
           first example, after the first retry, the call will succeed.
        
            >>> faux.calls[0][0] - faux.time
            5
            >>> time_flies(1) # 1 second
            0
            >>> deferred.called
            0
            >>> time_flies(1) # 1 second
            0
            >>> deferred.called
            0
            >>> time_flies(1) # 1 second
            0
            >>> deferred.called
            0
            >>> time_flies(1) # 1 second
            0
            >>> deferred.called
            0
            >>> time_flies(1) # 1 second
            1
            >>> deferred.called
            True
            >>> deferred.result
            1
            >>> len(faux.calls)
            0
        
           By default, the rescheduling backoff increases by five seconds for every
           retry, to a maximum of a 60 second backoff.
        
            >>> call = Partial(demo)
            >>> raise_error.extend([1] * 30)
            >>> len(faux.calls)
            0
            >>> deferred = call()
            >>> deferred.called
            0
            >>> len(faux.calls)
            1
            >>> def run(deferred):
            ...     sleeps = []
            ...     for i in range(31):
            ...         if deferred.called:
            ...             break
            ...         else:
            ...             sleep = faux.calls[0][0] - faux.time
            ...             sleeps.append(sleep)
            ...             time_flies(sleep)
            ...     else:
            ...         print 'oops'
            ...     return sleeps
            ...
            >>> sleeps = run(deferred)
            >>> deferred.result
            2
            >>> len(sleeps)
            30
            >>> sleeps # doctest: +ELLIPSIS
            [5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 60, 60,..., 60, 60, 60, 60]
        
           The default backoff values can be changed by setting the instance attributes
           ``initial_backoff``, ``backoff_increment``, and ``max_backoff``.
        
            >>> call = Partial(demo)
            >>> call.initial_backoff
            5
            >>> call.backoff_increment
            5
            >>> call.max_backoff
            60
            >>> call.initial_backoff = 4
            >>> call.backoff_increment = 2
            >>> call.max_backoff = 21
            >>> raise_error.extend([1] * 30)
            >>> len(faux.calls)
            0
            >>> deferred = call()
            >>> deferred.called
            0
            >>> len(faux.calls)
            1
            >>> sleeps = run(deferred)
            >>> deferred.result
            3
            >>> len(sleeps)
            30
            >>> sleeps # doctest: +ELLIPSIS
            [4, 6, 8, 10, 12, 14, 16, 18, 20, 21, 21, 21, 21, 21, 21,..., 21, 21, 21]
        
           A ClientDisconnected error can also occur during transaction commit.
        
            >>> storage_class.load = original_load
        
            >>> from transaction import TransactionManager
            >>> old_commit = TransactionManager.commit
            >>> commit_count = 0
            >>> error = None
            >>> max = 2
            >>> raise_error = [1] # change state to active
            >>> def new_commit(self):
            ...     if raise_error:
            ...         raise_error.pop()
            ...         raise ClientDisconnected()
            ...     else:
            ...         old_commit(self) # changing state to "active" or similar
            ...
            >>> TransactionManager.commit = new_commit
            >>> call = Partial(demo)
            >>> len(faux.calls)
            0
            >>> deferred = call()
            >>> deferred.called
            0
            >>> len(faux.calls)
            1
        
            >>> faux.calls[0][0] - faux.time
            5
            >>> time_flies(4) # 4 seconds
            0
            >>> deferred.called
            0
            >>> time_flies(1) # 1 second
            1
            >>> deferred.called
            True
            >>> deferred.result
            4
            >>> len(faux.calls)
            0
        
            >>> TransactionManager.commit = old_commit
            >>> _ = transaction.begin()
        
        .. [#show_sanitation] Before pickling, the failure includes full information
            about before and after the exception was caught, as well as locals and
            globals.  Everything has been repr'd, though, and the traceback object
            removed.
        
            >>> print res.getTraceback() # doctest: +ELLIPSIS
            Traceback (most recent call last):
              File ".../zc/twist/__init__.py", line ..., in __call__...
              File ".../twisted/internet/defer.py", line ..., in addCallback...
            --- <exception caught here> ---
              File ".../zc/twist/__init__.py", line ..., in _call...
              File "<doctest README.txt[...]>", line ..., in __call__...
            exceptions.TypeError: unsupported operand type(s) for +=: 'int' and 'str'
            <BLANKLINE>
        
            (The failure traceback at "verbose" detail is wildly verbose--this example
            takes out more than 90% of the text, just so you know.)
        
            >>> print res.getTraceback(detail='verbose') # doctest: +ELLIPSIS
            *--- Failure #... (pickled) ---
            .../zc/twist/__init__.py:...: __call__(...)
             [ Locals ]...
              args : "('I do not add well with integers',)...
             ( Globals )...
              Partial : "<class 'zc.twist.Partial'>...
            .../twisted/internet/defer.py:...: addCallback(...)
             [ Locals ]...
              args : '(<Deferred at ...
             ( Globals )...
              Deferred : '<class twisted.internet.defer.Deferred at ...
            --- <exception caught here> ---
            .../zc/twist/__init__.py:...: _call(...)
             [ Locals ]...
              args : "['I do not add well with integers']...
             ( Globals )...
              Partial : "<class 'zc.twist.Partial'>...
            <doctest README.txt[...]>:...: __call__(...)
             [ Locals ]...
              amount : "'I do not add well with integers'...
             ( Globals )...
              Partial : "<class 'zc.twist.Partial'>...
            exceptions.TypeError: unsupported operand type(s) for +=: 'int' and 'str'
            *--- End of Failure #... ---
            <BLANKLINE>
        
            After pickling, the failure only includes information for when the
            exception was caught and beyond (after the "--- <exception caught
            here> ---" lines above), does not have globals, and has local reprs
            truncated to a maximum of 20 characters plus "[...]" to indicate the
            truncation. This addresses past problems of large pickle size for
            failures, which can cause performance problems.
        
            >>> import pickle
            >>> print pickle.loads(pickle.dumps(res)).getTraceback()
            ... # doctest: +ELLIPSIS
            Traceback (most recent call last):
              File ".../zc/twist/__init__.py", line ..., in _call
                res = call(*args, **kwargs)
              File "<doctest README.txt[...]>", line ..., in __call__
                self.count += amount
            exceptions.TypeError: unsupported operand type(s) for +=: 'int' and 'str'
            <BLANKLINE>
        
            >>> print pickle.loads(pickle.dumps(res)).getTraceback(detail='verbose')
            ... # doctest: +ELLIPSIS
            *--- Failure #... (pickled) ---
            .../src/zc/twist/__init__.py:...: _call(...)
             [ Locals...
              self : '<zc.twist.Partial obj[...]...
             ( Globals )
            <doctest README.txt[...]>:...: __call__(...)
             [ Locals...
              amount : "'I do not add well wi[...]...
             ( Globals )
            exceptions.TypeError: unsupported operand type(s) for +=: 'int' and 'str'
            *--- End of Failure #... ---
            <BLANKLINE>
        
            In some cases, it is possible that a Failure object may include references
            to itself, for example, indirectly through a zc.async job whose result
            is a Failure.  Failure's __getstate__ method used to use deepcopy, which
            in cases like this could result in infinite recursion.
        
            >>> import zc.twist
            >>> class Kaboom(Exception):
            ...     pass
            >>> class Foo(object):
            ...     failure = None
            ...     def fail(self):
            ...         raise Kaboom, self
            >>> foo = Foo()
            >>> try:
            ...     foo.fail()
            ... except Kaboom:
            ...     foo.failure = zc.twist.Failure()
            >>> ignored = foo.failure.__getstate__() # used to cause RunTimeError.
        
        .. [#teardown_monkeypatch]
        
            >>> twisted.internet.reactor.callLater = oldCallLater
        
        .. [#setReactor]
        
            >>> db.setPoolSize(1)
            >>> db.getPoolSize()
            1
            >>> demo.count = 0
            >>> transaction.commit()
            >>> call = Partial(demo).setReactor(faux)
            >>> res = None
            >>> deferred = call()
            >>> d = deferred.addCallback(get_result)
            >>> call.attempt_count
            0
            >>> time_flies(.125) >= 1 # returns number of connection attempts
            True
            >>> call.attempt_count
            0
            >>> res # None
            >>> db.setPoolSize(2)
            >>> db.getPoolSize()
            2
            >>> time_flies(.25) >= 1
            True
            >>> call.attempt_count > 0
            True
            >>> res
            1
            >>> t = transaction.begin()
            >>> demo.count
            1
        
        If it takes more than a second or two, it will eventually just decide to grab
        one.  This behavior may change.
        
            >>> db.setPoolSize(1)
            >>> db.getPoolSize()
            1
            >>> call = Partial(demo).setReactor(faux)
            >>> res = None
            >>> deferred = call()
            >>> d = deferred.addCallback(get_result)
            >>> call.attempt_count
            0
            >>> time_flies(.125) >= 1
            True
            >>> call.attempt_count
            0
            >>> res # None
            >>> time_flies(2) >= 2 # for a total of at least 3
            True
            >>> res
            2
            >>> t = transaction.begin()
            >>> demo.count
            2
        
        
        =======
        Changes
        =======
        
        1.3.1 (2009-11-15)
        ------------------
        
        * Added missing import of twisted.python.failure.
        
        * Use db.pool with ZODB >= 3.9 and db._pools with ZODB < 3.9.
          The tests now pass with ZODB 3.9.
        
        * Tests pass in Python 2.6.
        
        1.3 (2008-06-19)
        ----------------
        
        * Handle ZEO.Exceptions.ClientDisconnected errors: retry forever, with a
          backoff, defaulting to a max backoff of 60 seconds.
        
        * Make number of times that ConflictErrors are retried configurable.
        
        1.2 (2008-04-09)
        ----------------
        
        * New subclass of twisted.python.failure.Failure begins with only reprs,
          and it pickles to exclude the stack, exclude the global vars in the frames,
          and truncate the reprs of the local vars in the frames.  The goal is to
          keep the pickle size of Failures down to a manageable size.  ``sanitize``
          now uses this class.
        
        1.1 (2008-03-27)
        ----------------
        
        * Now depends on twisted 8.0.1 or higher, which is newly setuptools
          compatible.  The twisted build is a little frightening, at least with
          Py2.4, with multiple warnings and errors, but works.  The dependency
          change is the reason for the bump to 1.1; this release has no new
          features.
        
        * setup.py uses os.path
        
        * C extension uses older comment style and has less cruft.
        
        1.0.1 (2008-03-14)
        ------------------
        
        * Bugfix: if you passed a slot method like a BTree.__setitem__, bad things
          would happen.
        
        1.0.0 (2008-03-13)
        ------------------
        
        * Add ability to specify an alternate reactor
        
        * Use bootstrap external
        
        0.1.1 (?)
        ---------
        
        * use zc.twisted, not twisted in setup.py, until twisted is setuptools-friendly
        
        0.1 (?)
        -------
        
        * Initial release-ish
        
        
        
Platform: UNKNOWN
