# -*- test-case-name: twisted.test.test_defer,twisted.test.test_defgen,twisted.internet.test.test_inlinecb -*- # Copyright (c) Twisted Matrix Laboratories. # See LICENSE for details. """ Support for results that aren't immediately available. Maintainer: Glyph Lefkowitz @var _NO_RESULT: The result used to represent the fact that there is no result. B{Never ever ever use this as an actual result for a Deferred}. You have been warned. @var _CONTINUE: A marker left in L{Deferred.callbacks} to indicate a Deferred chain. Always accompanied by a Deferred instance in the args tuple pointing at the Deferred which is chained to the Deferred which has this marker. """ from __future__ import division, absolute_import import traceback import types import warnings from sys import exc_info from functools import wraps # Twisted imports from twisted.python.compat import cmp, comparable from twisted.python import lockfile, log, failure from twisted.python.deprecate import warnAboutFunction class AlreadyCalledError(Exception): pass class CancelledError(Exception): """ This error is raised by default when a L{Deferred} is cancelled. """ class TimeoutError(Exception): """ This exception is deprecated. It is used only by the deprecated L{Deferred.setTimeout} method. """ def logError(err): log.err(err) return err def succeed(result): """ Return a L{Deferred} that has already had C{.callback(result)} called. This is useful when you're writing synchronous code to an asynchronous interface: i.e., some code is calling you expecting a L{Deferred} result, but you don't actually need to do anything asynchronous. Just return C{defer.succeed(theResult)}. See L{fail} for a version of this function that uses a failing L{Deferred} rather than a successful one. @param result: The result to give to the Deferred's 'callback' method. @rtype: L{Deferred} """ d = Deferred() d.callback(result) return d def fail(result=None): """ Return a L{Deferred} that has already had C{.errback(result)} called. See L{succeed}'s docstring for rationale. @param result: The same argument that L{Deferred.errback} takes. @raise NoCurrentExceptionError: If C{result} is C{None} but there is no current exception state. @rtype: L{Deferred} """ d = Deferred() d.errback(result) return d def execute(callable, *args, **kw): """ Create a L{Deferred} from a callable and arguments. Call the given function with the given arguments. Return a L{Deferred} which has been fired with its callback as the result of that invocation or its C{errback} with a L{Failure} for the exception thrown. """ try: result = callable(*args, **kw) except: return fail() else: return succeed(result) def maybeDeferred(f, *args, **kw): """ Invoke a function that may or may not return a L{Deferred}. Call the given function with the given arguments. If the returned object is a L{Deferred}, return it. If the returned object is a L{Failure}, wrap it with L{fail} and return it. Otherwise, wrap it in L{succeed} and return it. If an exception is raised, convert it to a L{Failure}, wrap it in L{fail}, and then return it. @type f: Any callable @param f: The callable to invoke @param args: The arguments to pass to C{f} @param kw: The keyword arguments to pass to C{f} @rtype: L{Deferred} @return: The result of the function call, wrapped in a L{Deferred} if necessary. """ try: result = f(*args, **kw) except: return fail(failure.Failure(captureVars=Deferred.debug)) if isinstance(result, Deferred): return result elif isinstance(result, failure.Failure): return fail(result) else: return succeed(result) def timeout(deferred): deferred.errback(failure.Failure(TimeoutError("Callback timed out"))) def passthru(arg): return arg def setDebugging(on): """ Enable or disable L{Deferred} debugging. When debugging is on, the call stacks from creation and invocation are recorded, and added to any L{AlreadyCalledErrors} we raise. """ Deferred.debug=bool(on) def getDebugging(): """ Determine whether L{Deferred} debugging is enabled. """ return Deferred.debug # See module docstring. _NO_RESULT = object() _CONTINUE = object() class Deferred: """ This is a callback which will be put off until later. Why do we want this? Well, in cases where a function in a threaded program would block until it gets a result, for Twisted it should not block. Instead, it should return a L{Deferred}. This can be implemented for protocols that run over the network by writing an asynchronous protocol for L{twisted.internet}. For methods that come from outside packages that are not under our control, we use threads (see for example L{twisted.enterprise.adbapi}). For more information about Deferreds, see doc/core/howto/defer.html or U{http://twistedmatrix.com/documents/current/core/howto/defer.html} When creating a Deferred, you may provide a canceller function, which will be called by d.cancel() to let you do any clean-up necessary if the user decides not to wait for the deferred to complete. @ivar called: A flag which is C{False} until either C{callback} or C{errback} is called and afterwards always C{True}. @type called: C{bool} @ivar paused: A counter of how many unmatched C{pause} calls have been made on this instance. @type paused: C{int} @ivar _suppressAlreadyCalled: A flag used by the cancellation mechanism which is C{True} if the Deferred has no canceller and has been cancelled, C{False} otherwise. If C{True}, it can be expected that C{callback} or C{errback} will eventually be called and the result should be silently discarded. @type _suppressAlreadyCalled: C{bool} @ivar _runningCallbacks: A flag which is C{True} while this instance is executing its callback chain, used to stop recursive execution of L{_runCallbacks} @type _runningCallbacks: C{bool} @ivar _chainedTo: If this Deferred is waiting for the result of another Deferred, this is a reference to the other Deferred. Otherwise, C{None}. """ called = False paused = 0 _debugInfo = None _suppressAlreadyCalled = False # Are we currently running a user-installed callback? Meant to prevent # recursive running of callbacks when a reentrant call to add a callback is # used. _runningCallbacks = False # Keep this class attribute for now, for compatibility with code that # sets it directly. debug = False _chainedTo = None def __init__(self, canceller=None): """ Initialize a L{Deferred}. @param canceller: a callable used to stop the pending operation scheduled by this L{Deferred} when L{Deferred.cancel} is invoked. The canceller will be passed the deferred whose cancelation is requested (i.e., self). If a canceller is not given, or does not invoke its argument's C{callback} or C{errback} method, L{Deferred.cancel} will invoke L{Deferred.errback} with a L{CancelledError}. Note that if a canceller is not given, C{callback} or C{errback} may still be invoked exactly once, even though defer.py will have already invoked C{errback}, as described above. This allows clients of code which returns a L{Deferred} to cancel it without requiring the L{Deferred} instantiator to provide any specific implementation support for cancellation. New in 10.1. @type canceller: a 1-argument callable which takes a L{Deferred}. The return result is ignored. """ self.callbacks = [] self._canceller = canceller if self.debug: self._debugInfo = DebugInfo() self._debugInfo.creator = traceback.format_stack()[:-1] def addCallbacks(self, callback, errback=None, callbackArgs=None, callbackKeywords=None, errbackArgs=None, errbackKeywords=None): """ Add a pair of callbacks (success and error) to this L{Deferred}. These will be executed when the 'master' callback is run. @return: C{self}. @rtype: a L{Deferred} """ assert callable(callback) assert errback == None or callable(errback) cbs = ((callback, callbackArgs, callbackKeywords), (errback or (passthru), errbackArgs, errbackKeywords)) self.callbacks.append(cbs) if self.called: self._runCallbacks() return self def addCallback(self, callback, *args, **kw): """ Convenience method for adding just a callback. See L{addCallbacks}. """ return self.addCallbacks(callback, callbackArgs=args, callbackKeywords=kw) def addErrback(self, errback, *args, **kw): """ Convenience method for adding just an errback. See L{addCallbacks}. """ return self.addCallbacks(passthru, errback, errbackArgs=args, errbackKeywords=kw) def addBoth(self, callback, *args, **kw): """ Convenience method for adding a single callable as both a callback and an errback. See L{addCallbacks}. """ return self.addCallbacks(callback, callback, callbackArgs=args, errbackArgs=args, callbackKeywords=kw, errbackKeywords=kw) def chainDeferred(self, d): """ Chain another L{Deferred} to this L{Deferred}. This method adds callbacks to this L{Deferred} to call C{d}'s callback or errback, as appropriate. It is merely a shorthand way of performing the following:: self.addCallbacks(d.callback, d.errback) When you chain a deferred d2 to another deferred d1 with d1.chainDeferred(d2), you are making d2 participate in the callback chain of d1. Thus any event that fires d1 will also fire d2. However, the converse is B{not} true; if d2 is fired d1 will not be affected. Note that unlike the case where chaining is caused by a L{Deferred} being returned from a callback, it is possible to cause the call stack size limit to be exceeded by chaining many L{Deferred}s together with C{chainDeferred}. @return: C{self}. @rtype: a L{Deferred} """ d._chainedTo = self return self.addCallbacks(d.callback, d.errback) def callback(self, result): """ Run all success callbacks that have been added to this L{Deferred}. Each callback will have its result passed as the first argument to the next; this way, the callbacks act as a 'processing chain'. If the success-callback returns a L{Failure} or raises an L{Exception}, processing will continue on the *error* callback chain. If a callback (or errback) returns another L{Deferred}, this L{Deferred} will be chained to it (and further callbacks will not run until that L{Deferred} has a result). An instance of L{Deferred} may only have either L{callback} or L{errback} called on it, and only once. @param result: The object which will be passed to the first callback added to this L{Deferred} (via L{addCallback}). @raise AlreadyCalledError: If L{callback} or L{errback} has already been called on this L{Deferred}. """ assert not isinstance(result, Deferred) self._startRunCallbacks(result) def errback(self, fail=None): """ Run all error callbacks that have been added to this L{Deferred}. Each callback will have its result passed as the first argument to the next; this way, the callbacks act as a 'processing chain'. Also, if the error-callback returns a non-Failure or doesn't raise an L{Exception}, processing will continue on the *success*-callback chain. If the argument that's passed to me is not a L{failure.Failure} instance, it will be embedded in one. If no argument is passed, a L{failure.Failure} instance will be created based on the current traceback stack. Passing a string as `fail' is deprecated, and will be punished with a warning message. An instance of L{Deferred} may only have either L{callback} or L{errback} called on it, and only once. @param fail: The L{Failure} object which will be passed to the first errback added to this L{Deferred} (via L{addErrback}). Alternatively, a L{Exception} instance from which a L{Failure} will be constructed (with no traceback) or C{None} to create a L{Failure} instance from the current exception state (with a traceback). @raise AlreadyCalledError: If L{callback} or L{errback} has already been called on this L{Deferred}. @raise NoCurrentExceptionError: If C{fail} is C{None} but there is no current exception state. """ if fail is None: fail = failure.Failure(captureVars=self.debug) elif not isinstance(fail, failure.Failure): fail = failure.Failure(fail) self._startRunCallbacks(fail) def pause(self): """ Stop processing on a L{Deferred} until L{unpause}() is called. """ self.paused = self.paused + 1 def unpause(self): """ Process all callbacks made since L{pause}() was called. """ self.paused = self.paused - 1 if self.paused: return if self.called: self._runCallbacks() def cancel(self): """ Cancel this L{Deferred}. If the L{Deferred} has not yet had its C{errback} or C{callback} method invoked, call the canceller function provided to the constructor. If that function does not invoke C{callback} or C{errback}, or if no canceller function was provided, errback with L{CancelledError}. If this L{Deferred} is waiting on another L{Deferred}, forward the cancellation to the other L{Deferred}. """ if not self.called: canceller = self._canceller if canceller: canceller(self) else: # Arrange to eat the callback that will eventually be fired # since there was no real canceller. self._suppressAlreadyCalled = True if not self.called: # There was no canceller, or the canceller didn't call # callback or errback. self.errback(failure.Failure(CancelledError())) elif isinstance(self.result, Deferred): # Waiting for another deferred -- cancel it instead. self.result.cancel() def _startRunCallbacks(self, result): if self.called: if self._suppressAlreadyCalled: self._suppressAlreadyCalled = False return if self.debug: if self._debugInfo is None: self._debugInfo = DebugInfo() extra = "\n" + self._debugInfo._getDebugTracebacks() raise AlreadyCalledError(extra) raise AlreadyCalledError if self.debug: if self._debugInfo is None: self._debugInfo = DebugInfo() self._debugInfo.invoker = traceback.format_stack()[:-2] self.called = True self.result = result self._runCallbacks() def _continuation(self): """ Build a tuple of callback and errback with L{_continue} to be used by L{_addContinue} and L{_removeContinue} on another Deferred. """ return ((_CONTINUE, (self,), None), (_CONTINUE, (self,), None)) def _runCallbacks(self): """ Run the chain of callbacks once a result is available. This consists of a simple loop over all of the callbacks, calling each with the current result and making the current result equal to the return value (or raised exception) of that call. If C{self._runningCallbacks} is true, this loop won't run at all, since it is already running above us on the call stack. If C{self.paused} is true, the loop also won't run, because that's what it means to be paused. The loop will terminate before processing all of the callbacks if a C{Deferred} without a result is encountered. If a C{Deferred} I{with} a result is encountered, that result is taken and the loop proceeds. @note: The implementation is complicated slightly by the fact that chaining (associating two Deferreds with each other such that one will wait for the result of the other, as happens when a Deferred is returned from a callback on another Deferred) is supported iteratively rather than recursively, to avoid running out of stack frames when processing long chains. """ if self._runningCallbacks: # Don't recursively run callbacks return # Keep track of all the Deferreds encountered while propagating results # up a chain. The way a Deferred gets onto this stack is by having # added its _continuation() to the callbacks list of a second Deferred # and then that second Deferred being fired. ie, if ever had _chainedTo # set to something other than None, you might end up on this stack. chain = [self] while chain: current = chain[-1] if current.paused: # This Deferred isn't going to produce a result at all. All the # Deferreds up the chain waiting on it will just have to... # wait. return finished = True current._chainedTo = None while current.callbacks: item = current.callbacks.pop(0) callback, args, kw = item[ isinstance(current.result, failure.Failure)] args = args or () kw = kw or {} # Avoid recursion if we can. if callback is _CONTINUE: # Give the waiting Deferred our current result and then # forget about that result ourselves. chainee = args[0] chainee.result = current.result current.result = None # Making sure to update _debugInfo if current._debugInfo is not None: current._debugInfo.failResult = None chainee.paused -= 1 chain.append(chainee) # Delay cleaning this Deferred and popping it from the chain # until after we've dealt with chainee. finished = False break try: current._runningCallbacks = True try: current.result = callback(current.result, *args, **kw) if current.result is current: warnAboutFunction( callback, "Callback returned the Deferred " "it was attached to; this breaks the " "callback chain and will raise an " "exception in the future.") finally: current._runningCallbacks = False except: # Including full frame information in the Failure is quite # expensive, so we avoid it unless self.debug is set. current.result = failure.Failure(captureVars=self.debug) else: if isinstance(current.result, Deferred): # The result is another Deferred. If it has a result, # we can take it and keep going. resultResult = getattr(current.result, 'result', _NO_RESULT) if resultResult is _NO_RESULT or isinstance(resultResult, Deferred) or current.result.paused: # Nope, it didn't. Pause and chain. current.pause() current._chainedTo = current.result # Note: current.result has no result, so it's not # running its callbacks right now. Therefore we can # append to the callbacks list directly instead of # using addCallbacks. current.result.callbacks.append(current._continuation()) break else: # Yep, it did. Steal it. current.result.result = None # Make sure _debugInfo's failure state is updated. if current.result._debugInfo is not None: current.result._debugInfo.failResult = None current.result = resultResult if finished: # As much of the callback chain - perhaps all of it - as can be # processed right now has been. The current Deferred is waiting on # another Deferred or for more callbacks. Before finishing with it, # make sure its _debugInfo is in the proper state. if isinstance(current.result, failure.Failure): # Stash the Failure in the _debugInfo for unhandled error # reporting. current.result.cleanFailure() if current._debugInfo is None: current._debugInfo = DebugInfo() current._debugInfo.failResult = current.result else: # Clear out any Failure in the _debugInfo, since the result # is no longer a Failure. if current._debugInfo is not None: current._debugInfo.failResult = None # This Deferred is done, pop it from the chain and move back up # to the Deferred which supplied us with our result. chain.pop() def __str__(self): """ Return a string representation of this C{Deferred}. """ cname = self.__class__.__name__ result = getattr(self, 'result', _NO_RESULT) myID = id(self) if self._chainedTo is not None: result = ' waiting on Deferred at 0x%x' % (id(self._chainedTo),) elif result is _NO_RESULT: result = '' else: result = ' current result: %r' % (result,) return "<%s at 0x%x%s>" % (cname, myID, result) __repr__ = __str__ class DebugInfo: """ Deferred debug helper. """ failResult = None def _getDebugTracebacks(self): info = '' if hasattr(self, "creator"): info += " C: Deferred was created:\n C:" info += "".join(self.creator).rstrip().replace("\n","\n C:") info += "\n" if hasattr(self, "invoker"): info += " I: First Invoker was:\n I:" info += "".join(self.invoker).rstrip().replace("\n","\n I:") info += "\n" return info def __del__(self): """ Print tracebacks and die. If the *last* (and I do mean *last*) callback leaves me in an error state, print a traceback (if said errback is a L{Failure}). """ if self.failResult is not None: log.msg("Unhandled error in Deferred:", isError=True) debugInfo = self._getDebugTracebacks() if debugInfo != '': log.msg("(debug: " + debugInfo + ")", isError=True) log.err(self.failResult) @comparable class FirstError(Exception): """ First error to occur in a L{DeferredList} if C{fireOnOneErrback} is set. @ivar subFailure: The L{Failure} that occurred. @type subFailure: L{Failure} @ivar index: The index of the L{Deferred} in the L{DeferredList} where it happened. @type index: C{int} """ def __init__(self, failure, index): Exception.__init__(self, failure, index) self.subFailure = failure self.index = index def __repr__(self): """ The I{repr} of L{FirstError} instances includes the repr of the wrapped failure's exception and the index of the L{FirstError}. """ return 'FirstError[#%d, %r]' % (self.index, self.subFailure.value) def __str__(self): """ The I{str} of L{FirstError} instances includes the I{str} of the entire wrapped failure (including its traceback and exception) and the index of the L{FirstError}. """ return 'FirstError[#%d, %s]' % (self.index, self.subFailure) def __cmp__(self, other): """ Comparison between L{FirstError} and other L{FirstError} instances is defined as the comparison of the index and sub-failure of each instance. L{FirstError} instances don't compare equal to anything that isn't a L{FirstError} instance. @since: 8.2 """ if isinstance(other, FirstError): return cmp( (self.index, self.subFailure), (other.index, other.subFailure)) return -1 class DeferredList(Deferred): """ L{DeferredList} is a tool for collecting the results of several Deferreds. This tracks a list of L{Deferred}s for their results, and makes a single callback when they have all completed. By default, the ultimate result is a list of (success, result) tuples, 'success' being a boolean. L{DeferredList} exposes the same API that L{Deferred} does, so callbacks and errbacks can be added to it in the same way. L{DeferredList} is implemented by adding callbacks and errbacks to each L{Deferred} in the list passed to it. This means callbacks and errbacks added to the Deferreds before they are passed to L{DeferredList} will change the result that L{DeferredList} sees (i.e., L{DeferredList} is not special). Callbacks and errbacks can also be added to the Deferreds after they are passed to L{DeferredList} and L{DeferredList} may change the result that they see. See the documentation for the C{__init__} arguments for more information. @ivar _deferredList: The C{list} of L{Deferred}s to track. """ fireOnOneCallback = False fireOnOneErrback = False def __init__(self, deferredList, fireOnOneCallback=False, fireOnOneErrback=False, consumeErrors=False): """ Initialize a DeferredList. @param deferredList: The list of deferreds to track. @type deferredList: C{list} of L{Deferred}s @param fireOnOneCallback: (keyword param) a flag indicating that this L{DeferredList} will fire when the first L{Deferred} in C{deferredList} fires with a non-failure result without waiting for any of the other Deferreds. When this flag is set, the DeferredList will fire with a two-tuple: the first element is the result of the Deferred which fired; the second element is the index in C{deferredList} of that Deferred. @type fireOnOneCallback: C{bool} @param fireOnOneErrback: (keyword param) a flag indicating that this L{DeferredList} will fire when the first L{Deferred} in C{deferredList} fires with a failure result without waiting for any of the other Deferreds. When this flag is set, if a Deferred in the list errbacks, the DeferredList will errback with a L{FirstError} failure wrapping the failure of that Deferred. @type fireOnOneErrback: C{bool} @param consumeErrors: (keyword param) a flag indicating that failures in any of the included L{Deferreds} should not be propagated to errbacks added to the individual L{Deferreds} after this L{DeferredList} is constructed. After constructing the L{DeferredList}, any errors in the individual L{Deferred}s will be converted to a callback result of C{None}. This is useful to prevent spurious 'Unhandled error in Deferred' messages from being logged. This does not prevent C{fireOnOneErrback} from working. @type consumeErrors: C{bool} """ self._deferredList = list(deferredList) self.resultList = [None] * len(self._deferredList) Deferred.__init__(self) if len(self._deferredList) == 0 and not fireOnOneCallback: self.callback(self.resultList) # These flags need to be set *before* attaching callbacks to the # deferreds, because the callbacks use these flags, and will run # synchronously if any of the deferreds are already fired. self.fireOnOneCallback = fireOnOneCallback self.fireOnOneErrback = fireOnOneErrback self.consumeErrors = consumeErrors self.finishedCount = 0 index = 0 for deferred in self._deferredList: deferred.addCallbacks(self._cbDeferred, self._cbDeferred, callbackArgs=(index,SUCCESS), errbackArgs=(index,FAILURE)) index = index + 1 def _cbDeferred(self, result, index, succeeded): """ (internal) Callback for when one of my deferreds fires. """ self.resultList[index] = (succeeded, result) self.finishedCount += 1 if not self.called: if succeeded == SUCCESS and self.fireOnOneCallback: self.callback((result, index)) elif succeeded == FAILURE and self.fireOnOneErrback: self.errback(failure.Failure(FirstError(result, index))) elif self.finishedCount == len(self.resultList): self.callback(self.resultList) if succeeded == FAILURE and self.consumeErrors: result = None return result def cancel(self): """ Cancel this L{DeferredList}. If the L{DeferredList} hasn't fired yet, cancel every L{Deferred} in the list. If the L{DeferredList} has fired, including the case where the C{fireOnOneCallback}/C{fireOnOneErrback} flag is set and the L{DeferredList} fires because one L{Deferred} in the list fires with a non-failure/failure result, do nothing in the C{cancel} method. """ if not self.called: for deferred in self._deferredList: try: deferred.cancel() except: log.err( _why="Exception raised from user supplied canceller") def _parseDListResult(l, fireOnOneErrback=False): if __debug__: for success, value in l: assert success return [x[1] for x in l] def gatherResults(deferredList, consumeErrors=False): """ Returns, via a L{Deferred}, a list with the results of the given L{Deferred}s - in effect, a "join" of multiple deferred operations. The returned L{Deferred} will fire when I{all} of the provided L{Deferred}s have fired, or when any one of them has failed. This method can be cancelled by calling the C{cancel} method of the L{Deferred}, all the L{Deferred}s in the list will be cancelled. This differs from L{DeferredList} in that you don't need to parse the result for success/failure. @type deferredList: C{list} of L{Deferred}s @param consumeErrors: (keyword param) a flag, defaulting to False, indicating that failures in any of the given L{Deferreds} should not be propagated to errbacks added to the individual L{Deferreds} after this L{gatherResults} invocation. Any such errors in the individual L{Deferred}s will be converted to a callback result of C{None}. This is useful to prevent spurious 'Unhandled error in Deferred' messages from being logged. This parameter is available since 11.1.0. @type consumeErrors: C{bool} """ d = DeferredList(deferredList, fireOnOneErrback=True, consumeErrors=consumeErrors) d.addCallback(_parseDListResult) return d # Constants for use with DeferredList SUCCESS = True FAILURE = False ## deferredGenerator class waitForDeferred: """ See L{deferredGenerator}. """ def __init__(self, d): if not isinstance(d, Deferred): raise TypeError("You must give waitForDeferred a Deferred. You gave it %r." % (d,)) self.d = d def getResult(self): if isinstance(self.result, failure.Failure): self.result.raiseException() return self.result def _deferGenerator(g, deferred): """ See L{deferredGenerator}. """ result = None # This function is complicated by the need to prevent unbounded recursion # arising from repeatedly yielding immediately ready deferreds. This while # loop and the waiting variable solve that by manually unfolding the # recursion. waiting = [True, # defgen is waiting for result? None] # result while 1: try: result = next(g) except StopIteration: deferred.callback(result) return deferred except: deferred.errback() return deferred # Deferred.callback(Deferred) raises an error; we catch this case # early here and give a nicer error message to the user in case # they yield a Deferred. if isinstance(result, Deferred): return fail(TypeError("Yield waitForDeferred(d), not d!")) if isinstance(result, waitForDeferred): # a waitForDeferred was yielded, get the result. # Pass result in so it don't get changed going around the loop # This isn't a problem for waiting, as it's only reused if # gotResult has already been executed. def gotResult(r, result=result): result.result = r if waiting[0]: waiting[0] = False waiting[1] = r else: _deferGenerator(g, deferred) result.d.addBoth(gotResult) if waiting[0]: # Haven't called back yet, set flag so that we get reinvoked # and return from the loop waiting[0] = False return deferred # Reset waiting to initial values for next loop waiting[0] = True waiting[1] = None result = None def deferredGenerator(f): """ L{deferredGenerator} and L{waitForDeferred} help you write L{Deferred}-using code that looks like a regular sequential function. Consider the use of L{inlineCallbacks} instead, which can accomplish the same thing in a more concise manner. There are two important functions involved: L{waitForDeferred}, and L{deferredGenerator}. They are used together, like this:: @deferredGenerator def thingummy(): thing = waitForDeferred(makeSomeRequestResultingInDeferred()) yield thing thing = thing.getResult() print thing #the result! hoorj! L{waitForDeferred} returns something that you should immediately yield; when your generator is resumed, calling C{thing.getResult()} will either give you the result of the L{Deferred} if it was a success, or raise an exception if it was a failure. Calling C{getResult} is B{absolutely mandatory}. If you do not call it, I{your program will not work}. L{deferredGenerator} takes one of these waitForDeferred-using generator functions and converts it into a function that returns a L{Deferred}. The result of the L{Deferred} will be the last value that your generator yielded unless the last value is a L{waitForDeferred} instance, in which case the result will be C{None}. If the function raises an unhandled exception, the L{Deferred} will errback instead. Remember that C{return result} won't work; use C{yield result; return} in place of that. Note that not yielding anything from your generator will make the L{Deferred} result in C{None}. Yielding a L{Deferred} from your generator is also an error condition; always yield C{waitForDeferred(d)} instead. The L{Deferred} returned from your deferred generator may also errback if your generator raised an exception. For example:: @deferredGenerator def thingummy(): thing = waitForDeferred(makeSomeRequestResultingInDeferred()) yield thing thing = thing.getResult() if thing == 'I love Twisted': # will become the result of the Deferred yield 'TWISTED IS GREAT!' return else: # will trigger an errback raise Exception('DESTROY ALL LIFE') Put succinctly, these functions connect deferred-using code with this 'fake blocking' style in both directions: L{waitForDeferred} converts from a L{Deferred} to the 'blocking' style, and L{deferredGenerator} converts from the 'blocking' style to a L{Deferred}. """ @wraps(f) def unwindGenerator(*args, **kwargs): return _deferGenerator(f(*args, **kwargs), Deferred()) return unwindGenerator ## inlineCallbacks class _DefGen_Return(BaseException): def __init__(self, value): self.value = value def returnValue(val): """ Return val from a L{inlineCallbacks} generator. Note: this is currently implemented by raising an exception derived from L{BaseException}. You might want to change any 'except:' clauses to an 'except Exception:' clause so as not to catch this exception. Also: while this function currently will work when called from within arbitrary functions called from within the generator, do not rely upon this behavior. """ raise _DefGen_Return(val) def _inlineCallbacks(result, g, deferred): """ See L{inlineCallbacks}. """ # This function is complicated by the need to prevent unbounded recursion # arising from repeatedly yielding immediately ready deferreds. This while # loop and the waiting variable solve that by manually unfolding the # recursion. waiting = [True, # waiting for result? None] # result while 1: try: # Send the last result back as the result of the yield expression. isFailure = isinstance(result, failure.Failure) if isFailure: result = result.throwExceptionIntoGenerator(g) else: result = g.send(result) except StopIteration: # fell off the end, or "return" statement deferred.callback(None) return deferred except _DefGen_Return as e: # returnValue() was called; time to give a result to the original # Deferred. First though, let's try to identify the potentially # confusing situation which results when returnValue() is # accidentally invoked from a different function, one that wasn't # decorated with @inlineCallbacks. # The traceback starts in this frame (the one for # _inlineCallbacks); the next one down should be the application # code. appCodeTrace = exc_info()[2].tb_next if isFailure: # If we invoked this generator frame by throwing an exception # into it, then throwExceptionIntoGenerator will consume an # additional stack frame itself, so we need to skip that too. appCodeTrace = appCodeTrace.tb_next # Now that we've identified the frame being exited by the # exception, let's figure out if returnValue was called from it # directly. returnValue itself consumes a stack frame, so the # application code will have a tb_next, but it will *not* have a # second tb_next. if appCodeTrace.tb_next.tb_next: # If returnValue was invoked non-local to the frame which it is # exiting, identify the frame that ultimately invoked # returnValue so that we can warn the user, as this behavior is # confusing. ultimateTrace = appCodeTrace while ultimateTrace.tb_next.tb_next: ultimateTrace = ultimateTrace.tb_next filename = ultimateTrace.tb_frame.f_code.co_filename lineno = ultimateTrace.tb_lineno warnings.warn_explicit( "returnValue() in %r causing %r to exit: " "returnValue should only be invoked by functions decorated " "with inlineCallbacks" % ( ultimateTrace.tb_frame.f_code.co_name, appCodeTrace.tb_frame.f_code.co_name), DeprecationWarning, filename, lineno) deferred.callback(e.value) return deferred except: deferred.errback() return deferred if isinstance(result, Deferred): # a deferred was yielded, get the result. def gotResult(r): if waiting[0]: waiting[0] = False waiting[1] = r else: _inlineCallbacks(r, g, deferred) result.addBoth(gotResult) if waiting[0]: # Haven't called back yet, set flag so that we get reinvoked # and return from the loop waiting[0] = False return deferred result = waiting[1] # Reset waiting to initial values for next loop. gotResult uses # waiting, but this isn't a problem because gotResult is only # executed once, and if it hasn't been executed yet, the return # branch above would have been taken. waiting[0] = True waiting[1] = None return deferred def inlineCallbacks(f): """ inlineCallbacks helps you write L{Deferred}-using code that looks like a regular sequential function. For example:: @inlineCallBacks def thingummy(): thing = yield makeSomeRequestResultingInDeferred() print(thing) # the result! hoorj! When you call anything that results in a L{Deferred}, you can simply yield it; your generator will automatically be resumed when the Deferred's result is available. The generator will be sent the result of the L{Deferred} with the 'send' method on generators, or if the result was a failure, 'throw'. Things that are not L{Deferred}s may also be yielded, and your generator will be resumed with the same object sent back. This means C{yield} performs an operation roughly equivalent to L{maybeDeferred}. Your inlineCallbacks-enabled generator will return a L{Deferred} object, which will result in the return value of the generator (or will fail with a failure object if your generator raises an unhandled exception). Note that you can't use C{return result} to return a value; use C{returnValue(result)} instead. Falling off the end of the generator, or simply using C{return} will cause the L{Deferred} to have a result of C{None}. Be aware that L{returnValue} will not accept a L{Deferred} as a parameter. If you believe the thing you'd like to return could be a L{Deferred}, do this:: result = yield result returnValue(result) The L{Deferred} returned from your deferred generator may errback if your generator raised an exception:: @inlineCallbacks def thingummy(): thing = yield makeSomeRequestResultingInDeferred() if thing == 'I love Twisted': # will become the result of the Deferred returnValue('TWISTED IS GREAT!') else: # will trigger an errback raise Exception('DESTROY ALL LIFE') """ @wraps(f) def unwindGenerator(*args, **kwargs): try: gen = f(*args, **kwargs) except _DefGen_Return: raise TypeError( "inlineCallbacks requires %r to produce a generator; instead" "caught returnValue being used in a non-generator" % (f,)) if not isinstance(gen, types.GeneratorType): raise TypeError( "inlineCallbacks requires %r to produce a generator; " "instead got %r" % (f, gen)) return _inlineCallbacks(None, gen, Deferred()) return unwindGenerator ## DeferredLock/DeferredQueue class _ConcurrencyPrimitive(object): def __init__(self): self.waiting = [] def _releaseAndReturn(self, r): self.release() return r def run(*args, **kwargs): """ Acquire, run, release. This function takes a callable as its first argument and any number of other positional and keyword arguments. When the lock or semaphore is acquired, the callable will be invoked with those arguments. The callable may return a L{Deferred}; if it does, the lock or semaphore won't be released until that L{Deferred} fires. @return: L{Deferred} of function result. """ if len(args) < 2: if not args: raise TypeError("run() takes at least 2 arguments, none given.") raise TypeError("%s.run() takes at least 2 arguments, 1 given" % ( args[0].__class__.__name__,)) self, f = args[:2] args = args[2:] def execute(ignoredResult): d = maybeDeferred(f, *args, **kwargs) d.addBoth(self._releaseAndReturn) return d d = self.acquire() d.addCallback(execute) return d class DeferredLock(_ConcurrencyPrimitive): """ A lock for event driven systems. @ivar locked: C{True} when this Lock has been acquired, false at all other times. Do not change this value, but it is useful to examine for the equivalent of a "non-blocking" acquisition. """ locked = False def _cancelAcquire(self, d): """ Remove a deferred d from our waiting list, as the deferred has been canceled. Note: We do not need to wrap this in a try/except to catch d not being in self.waiting because this canceller will not be called if d has fired. release() pops a deferred out of self.waiting and calls it, so the canceller will no longer be called. @param d: The deferred that has been canceled. """ self.waiting.remove(d) def acquire(self): """ Attempt to acquire the lock. Returns a L{Deferred} that fires on lock acquisition with the L{DeferredLock} as the value. If the lock is locked, then the Deferred is placed at the end of a waiting list. @return: a L{Deferred} which fires on lock acquisition. @rtype: a L{Deferred} """ d = Deferred(canceller=self._cancelAcquire) if self.locked: self.waiting.append(d) else: self.locked = True d.callback(self) return d def release(self): """ Release the lock. If there is a waiting list, then the first L{Deferred} in that waiting list will be called back. Should be called by whomever did the L{acquire}() when the shared resource is free. """ assert self.locked, "Tried to release an unlocked lock" self.locked = False if self.waiting: # someone is waiting to acquire lock self.locked = True d = self.waiting.pop(0) d.callback(self) class DeferredSemaphore(_ConcurrencyPrimitive): """ A semaphore for event driven systems. If you are looking into this as a means of limiting parallelism, you might find L{twisted.internet.task.Cooperator} more useful. @ivar tokens: At most this many users may acquire this semaphore at once. @type tokens: C{int} @ivar limit: The difference between C{tokens} and the number of users which have currently acquired this semaphore. @type limit: C{int} """ def __init__(self, tokens): _ConcurrencyPrimitive.__init__(self) if tokens < 1: raise ValueError("DeferredSemaphore requires tokens >= 1") self.tokens = tokens self.limit = tokens def _cancelAcquire(self, d): """ Remove a deferred d from our waiting list, as the deferred has been canceled. Note: We do not need to wrap this in a try/except to catch d not being in self.waiting because this canceller will not be called if d has fired. release() pops a deferred out of self.waiting and calls it, so the canceller will no longer be called. @param d: The deferred that has been canceled. """ self.waiting.remove(d) def acquire(self): """ Attempt to acquire the token. @return: a L{Deferred} which fires on token acquisition. """ assert self.tokens >= 0, "Internal inconsistency?? tokens should never be negative" d = Deferred(canceller=self._cancelAcquire) if not self.tokens: self.waiting.append(d) else: self.tokens = self.tokens - 1 d.callback(self) return d def release(self): """ Release the token. Should be called by whoever did the L{acquire}() when the shared resource is free. """ assert self.tokens < self.limit, "Someone released me too many times: too many tokens!" self.tokens = self.tokens + 1 if self.waiting: # someone is waiting to acquire token self.tokens = self.tokens - 1 d = self.waiting.pop(0) d.callback(self) class QueueOverflow(Exception): pass class QueueUnderflow(Exception): pass class DeferredQueue(object): """ An event driven queue. Objects may be added as usual to this queue. When an attempt is made to retrieve an object when the queue is empty, a L{Deferred} is returned which will fire when an object becomes available. @ivar size: The maximum number of objects to allow into the queue at a time. When an attempt to add a new object would exceed this limit, L{QueueOverflow} is raised synchronously. C{None} for no limit. @ivar backlog: The maximum number of L{Deferred} gets to allow at one time. When an attempt is made to get an object which would exceed this limit, L{QueueUnderflow} is raised synchronously. C{None} for no limit. """ def __init__(self, size=None, backlog=None): self.waiting = [] self.pending = [] self.size = size self.backlog = backlog def _cancelGet(self, d): """ Remove a deferred d from our waiting list, as the deferred has been canceled. Note: We do not need to wrap this in a try/except to catch d not being in self.waiting because this canceller will not be called if d has fired. put() pops a deferred out of self.waiting and calls it, so the canceller will no longer be called. @param d: The deferred that has been canceled. """ self.waiting.remove(d) def put(self, obj): """ Add an object to this queue. @raise QueueOverflow: Too many objects are in this queue. """ if self.waiting: self.waiting.pop(0).callback(obj) elif self.size is None or len(self.pending) < self.size: self.pending.append(obj) else: raise QueueOverflow() def get(self): """ Attempt to retrieve and remove an object from the queue. @return: a L{Deferred} which fires with the next object available in the queue. @raise QueueUnderflow: Too many (more than C{backlog}) L{Deferred}s are already waiting for an object from this queue. """ if self.pending: return succeed(self.pending.pop(0)) elif self.backlog is None or len(self.waiting) < self.backlog: d = Deferred(canceller=self._cancelGet) self.waiting.append(d) return d else: raise QueueUnderflow() class AlreadyTryingToLockError(Exception): """ Raised when L{DeferredFilesystemLock.deferUntilLocked} is called twice on a single L{DeferredFilesystemLock}. """ class DeferredFilesystemLock(lockfile.FilesystemLock): """ A L{FilesystemLock} that allows for a L{Deferred} to be fired when the lock is acquired. @ivar _scheduler: The object in charge of scheduling retries. In this implementation this is parameterized for testing. @ivar _interval: The retry interval for an L{IReactorTime} based scheduler. @ivar _tryLockCall: A L{DelayedCall} based on C{_interval} that will manage the next retry for aquiring the lock. @ivar _timeoutCall: A L{DelayedCall} based on C{deferUntilLocked}'s timeout argument. This is in charge of timing out our attempt to acquire the lock. """ _interval = 1 _tryLockCall = None _timeoutCall = None def __init__(self, name, scheduler=None): """ @param name: The name of the lock to acquire @param scheduler: An object which provides L{IReactorTime} """ lockfile.FilesystemLock.__init__(self, name) if scheduler is None: from twisted.internet import reactor scheduler = reactor self._scheduler = scheduler def deferUntilLocked(self, timeout=None): """ Wait until we acquire this lock. This method is not safe for concurrent use. @type timeout: C{float} or C{int} @param timeout: the number of seconds after which to time out if the lock has not been acquired. @return: a L{Deferred} which will callback when the lock is acquired, or errback with a L{TimeoutError} after timing out or an L{AlreadyTryingToLockError} if the L{deferUntilLocked} has already been called and not successfully locked the file. """ if self._tryLockCall is not None: return fail( AlreadyTryingToLockError( "deferUntilLocked isn't safe for concurrent use.")) def _cancelLock(reason): """ Cancel a L{DeferredFilesystemLock.deferUntilLocked} call. @type reason: L{failure.Failure} @param reason: The reason why the call is cancelled. """ self._tryLockCall.cancel() self._tryLockCall = None if self._timeoutCall is not None and self._timeoutCall.active(): self._timeoutCall.cancel() self._timeoutCall = None if self.lock(): d.callback(None) else: d.errback(reason) d = Deferred(lambda deferred: _cancelLock(CancelledError())) def _tryLock(): if self.lock(): if self._timeoutCall is not None: self._timeoutCall.cancel() self._timeoutCall = None self._tryLockCall = None d.callback(None) else: if timeout is not None and self._timeoutCall is None: reason = failure.Failure(TimeoutError( "Timed out aquiring lock: %s after %fs" % ( self.name, timeout))) self._timeoutCall = self._scheduler.callLater( timeout, _cancelLock, reason) self._tryLockCall = self._scheduler.callLater( self._interval, _tryLock) _tryLock() return d __all__ = ["Deferred", "DeferredList", "succeed", "fail", "FAILURE", "SUCCESS", "AlreadyCalledError", "TimeoutError", "gatherResults", "maybeDeferred", "waitForDeferred", "deferredGenerator", "inlineCallbacks", "returnValue", "DeferredLock", "DeferredSemaphore", "DeferredQueue", "DeferredFilesystemLock", "AlreadyTryingToLockError", ]