Bug #788520: Queue class has logic error when non-blocking

I don't agree it had a bug (see the report), so this is *not* a candidate
for backporting, but the docs were confusing and the Queue implementation
was old enough to vote.

Rewrote put/put_nowait/get/get_nowait from scratch, to use a pair of
Conditions (not_full and not_empty), sharing a common mutex.  The code
is 1/4 the size now, and 6.25x easier to understand.  For blocking
with timeout, we also get to reuse (indirectly) the tedious timeout
code from threading.Condition.  The Full and Empty exceptions raised
by non-blocking calls are now easy (instead of nearly impossible) to
explain truthfully:  Full is raised if and only if the Queue truly
is full when the non-blocking put call checks the queue size, and
similarly for Empty versus non-blocking get.

What I don't know is whether the new implementation is slower (or
faster) than the old one.  I don't really care.  Anyone who cares
a lot is encouraged to check that.

Doc/lib/libqueue.tex

@@ -26,13 +26,13 @@ zero, the queue size is infinite.
 \begin{excdesc}{Empty}
 Exception raised when non-blocking \method{get()} (or
 \method{get_nowait()}) is called on a \class{Queue} object which is
-empty or locked.
+empty.
 \end{excdesc}
 
 \begin{excdesc}{Full}
 Exception raised when non-blocking \method{put()} (or
 \method{put_nowait()}) is called on a \class{Queue} object which is
-full or locked.
+full.
 \end{excdesc}
 
 \subsection{Queue Objects}
@@ -51,7 +51,7 @@ semantics, this number is not reliable.
 
 \begin{methoddesc}{empty}{}
 Return \code{True} if the queue is empty, \code{False} otherwise.
-Becauseof multithreading semantics, this is not reliable.
+Because of multithreading semantics, this is not reliable.
 \end{methoddesc}
 
 \begin{methoddesc}{full}{}

Lib/Queue.py

@@ -1,6 +1,6 @@
 """A multi-producer, multi-consumer queue."""
 
-from time import time as _time, sleep as _sleep
+from time import time as _time
 from collections import deque
 
 __all__ = ['Empty', 'Full', 'Queue']
@@ -20,14 +20,21 @@ class Queue:
         If maxsize is <= 0, the queue size is infinite.
         """
         try:
-            import thread
+            import threading
         except ImportError:
-            import dummy_thread as thread
+            import dummy_threading as threading
         self._init(maxsize)
-        self.mutex = thread.allocate_lock()
+        # mutex must be held whenever the queue is mutating.  All methods
-        self.esema = thread.allocate_lock()
+        # that acquire mutex must release it before returning.  mutex
-        self.esema.acquire()
+        # is shared between the two conditions, so acquiring and
-        self.fsema = thread.allocate_lock()
+        # releasing the conditions also acquires and releases mutex.
+        self.mutex = threading.Lock()
+        # Notify not_empty whenever an item is added to the queue; a
+        # thread waiting to get is notified then.
+        self.not_empty = threading.Condition(self.mutex)
+        # Notify not_full whenever an item is removed from the queue;
+        # a thread waiting to put is notified then.
+        self.not_full = threading.Condition(self.mutex)
 
     def qsize(self):
         """Return the approximate size of the queue (not reliable!)."""
@@ -61,51 +68,26 @@ class Queue:
         is immediately available, else raise the Full exception ('timeout'
         is ignored in that case).
         """
-        if block:
+        if not block:
-            if timeout is None:
+            return self.put_nowait(item)
-                # blocking, w/o timeout, i.e. forever
+        self.not_full.acquire()
-                self.fsema.acquire()
-            elif timeout >= 0:
-                # waiting max. 'timeout' seconds.
-                # this code snipped is from threading.py: _Event.wait():
-                # Balancing act:  We can't afford a pure busy loop, so we
-                # have to sleep; but if we sleep the whole timeout time,
-                # we'll be unresponsive.  The scheme here sleeps very
-                # little at first, longer as time goes on, but never longer
-                # than 20 times per second (or the timeout time remaining).
-                delay = 0.0005 # 500 us -> initial delay of 1 ms
-                endtime = _time() + timeout
-                while True:
-                    if self.fsema.acquire(0):
-                        break
-                    remaining = endtime - _time()
-                    if remaining <= 0:  #time is over and no slot was free
-                        raise Full
-                    delay = min(delay * 2, remaining, .05)
-                    _sleep(delay)       #reduce CPU usage by using a sleep
-            else:
-                raise ValueError("'timeout' must be a positive number")
-        elif not self.fsema.acquire(0):
-            raise Full
-        self.mutex.acquire()
-        release_fsema = True
         try:
-            was_empty = self._empty()
+            if timeout is None:
+                while self._full():
+                    self.not_full.wait()
+            else:
+                if timeout < 0:
+                    raise ValueError("'timeout' must be a positive number")
+                endtime = _time() + timeout
+                while self._full():
+                    remaining = endtime - _time()
+                    if remaining < 0.0:
+                        raise Full
+                    self.not_full.wait(remaining)
             self._put(item)
-            # If we fail before here, the empty state has
+            self.not_empty.notify()
-            # not changed, so we can skip the release of esema
-            if was_empty:
-                self.esema.release()
-            # If we fail before here, the queue can not be full, so
-            # release_full_sema remains True
-            release_fsema = not self._full()
         finally:
-            # Catching system level exceptions here (RecursionDepth,
+            self.not_full.release()
-            # OutOfMemory, etc) - so do as little as possible in terms
-            # of Python calls.
-            if release_fsema:
-                self.fsema.release()
-            self.mutex.release()
 
     def put_nowait(self, item):
         """Put an item into the queue without blocking.
@@ -113,7 +95,15 @@ class Queue:
         Only enqueue the item if a free slot is immediately available.
         Otherwise raise the Full exception.
         """
-        return self.put(item, False)
+        self.not_full.acquire()
+        try:
+            if self._full():
+                raise Full
+            else:
+                self._put(item)
+                self.not_empty.notify()
+        finally:
+            self.not_full.release()
 
     def get(self, block=True, timeout=None):
         """Remove and return an item from the queue.
@@ -126,49 +116,27 @@ class Queue:
         available, else raise the Empty exception ('timeout' is ignored
         in that case).
         """
-        if block:
+        if not block:
-            if timeout is None:
+            return self.get_nowait()
-                # blocking, w/o timeout, i.e. forever
+        self.not_empty.acquire()
-                self.esema.acquire()
-            elif timeout >= 0:
-                # waiting max. 'timeout' seconds.
-                # this code snipped is from threading.py: _Event.wait():
-                # Balancing act:  We can't afford a pure busy loop, so we
-                # have to sleep; but if we sleep the whole timeout time,
-                # we'll be unresponsive.  The scheme here sleeps very
-                # little at first, longer as time goes on, but never longer
-                # than 20 times per second (or the timeout time remaining).
-                delay = 0.0005 # 500 us -> initial delay of 1 ms
-                endtime = _time() + timeout
-                while 1:
-                    if self.esema.acquire(0):
-                        break
-                    remaining = endtime - _time()
-                    if remaining <= 0:  #time is over and no element arrived
-                        raise Empty
-                    delay = min(delay * 2, remaining, .05)
-                    _sleep(delay)       #reduce CPU usage by using a sleep
-            else:
-                raise ValueError("'timeout' must be a positive number")
-        elif not self.esema.acquire(0):
-            raise Empty
-        self.mutex.acquire()
-        release_esema = True
         try:
-            was_full = self._full()
+            if timeout is None:
+                while self._empty():
+                    self.not_empty.wait()
+            else:
+                if timeout < 0:
+                    raise ValueError("'timeout' must be a positive number")
+                endtime = _time() + timeout
+                while self._empty():
+                    remaining = endtime - _time()
+                    if remaining < 0.0:
+                        raise Empty
+                    self.not_empty.wait(remaining)
             item = self._get()
-            # If we fail before here, the full state has
+            self.not_full.notify()
-            # not changed, so we can skip the release of fsema
+            return item
-            if was_full:
-                self.fsema.release()
-            # Failure means empty state also unchanged - release_esema
-            # remains True.
-            release_esema = not self._empty()
         finally:
-            if release_esema:
+            self.not_empty.release()
-                self.esema.release()
-            self.mutex.release()
-        return item
 
     def get_nowait(self):
         """Remove and return an item from the queue without blocking.
@@ -176,7 +144,16 @@ class Queue:
         Only get an item if one is immediately available. Otherwise
         raise the Empty exception.
         """
-        return self.get(False)
+        self.not_empty.acquire()
+        try:
+            if self._empty():
+                raise Empty
+            else:
+                item = self._get()
+                self.not_full.notify()
+                return item
+        finally:
+            self.not_empty.release()
 
     # Override these methods to implement other queue organizations
     # (e.g. stack or priority queue).

Misc/NEWS

@@ -29,6 +29,17 @@ Extension modules
 Library
 -------
 
+- Bug #788520.  Queue.{get, get_nowait, put, put_nowait} have new
+  implementations, exploiting Conditions (which didn't exist at the time
+  Queue was introduced).  A minor semantic change is that the Full and
+  Empty exceptions raised by non-blocking calls now occur only if the
+  queue truly was full or empty at the instant the queue was checked (of
+  course the Queue may no longer be full or empty by the time a calling
+  thread sees those exceptions, though).  Before, the exceptions could
+  also be raised if it was "merely inconvenient" for the implementation
+  to determine the true state of the Queue (because the Queue was locked
+  by some other method in progress).
+
 - Bugs #979794 and #980117: difflib.get_grouped_opcodes() now handles the
   case of comparing two empty lists.  This affected both context_diff() and
   unified_diff(),