ENH Add a switch to override existing global variables

CHANGES.md

@@ -1,3 +1,12 @@
+0.7.0
+=====
+
+- Add a switch in `cloudpickle.dump` speficying if variables present in the
+  written pickle string should override colluding values present in their new
+  namespace at unpickling time.
+  ([issue #214](https://github.com/cloudpipe/cloudpickle/issues/214))
+
+
 0.6.1
 =====
 

cloudpickle/cloudpickle.py

@@ -266,16 +266,46 @@ def _walk_global_ops(code):
 
 
 class CloudPickler(Pickler):
+    """Pickler with extended pickling abilities.
+
+    Parameters
+    ----------
+    file :
+
+    protocol : int
+
+    override_globals : bool
+        Functions belonging to the same module (the __main__ or a dynamic
+        module) share the same global variables. Using cloudpickle, they will
+        share those global variables in the process they are unpickled in as
+        well. However, the state of those global variables may differ in the
+        pickled functions, for example if one global variable was mutated
+        between two calls to pickle.dump. override_globals is a
+        switch that allows two different behaviors:
+
+        * If override_globals=True, the global variables of the
+          module containing this function will be overridden in the process
+          which load the pickled object with the values of these variables at
+          the pickling time.
+        * If override_globals=False, the global variables of the
+          module containing the definition of the nested or dynamically defined
+          function will not be overridden in the process loading the pickled
+          object. If the global variable is not declared, it will be
+          initialized with the value at pickling time.
+
+    """
 
     dispatch = Pickler.dispatch.copy()
 
-    def __init__(self, file, protocol=None):
+    def __init__(self, file, protocol=None, override_globals=True):
         if protocol is None:
             protocol = DEFAULT_PROTOCOL
         Pickler.__init__(self, file, protocol=protocol)
         # map ids to dictionary. used to ensure that functions can share global env
         self.globals_ref = {}
 
+        self.override_globals = override_globals
+
     def dump(self, obj):
         self.inject_addons()
         try:
@@ -590,6 +620,7 @@ def save_function_tuple(self, func):
             'module': func.__module__,
             'name': func.__name__,
             'doc': func.__doc__,
+            'override_globals': self.override_globals
         }
         if hasattr(func, '__annotations__') and sys.version_info >= (3, 7):
             state['annotations'] = func.__annotations__
@@ -921,7 +952,7 @@ def _rebuild_tornado_coroutine(func):
 
 # Shorthands for legacy support
 
-def dump(obj, file, protocol=None):
+def dump(obj, file, protocol=None, override_globals=True):
     """Serialize obj as bytes streamed into file
 
     protocol defaults to cloudpickle.DEFAULT_PROTOCOL which is an alias to
@@ -931,10 +962,11 @@ def dump(obj, file, protocol=None):
     Set protocol=pickle.DEFAULT_PROTOCOL instead if you need to ensure
     compatibility with older versions of Python.
     """
-    CloudPickler(file, protocol=protocol).dump(obj)
+    CloudPickler(file, protocol=protocol,
+                 override_globals=override_globals).dump(obj)
 
 
-def dumps(obj, protocol=None):
+def dumps(obj, protocol=None, override_globals=True):
     """Serialize obj as a string of bytes allocated in memory
 
     protocol defaults to cloudpickle.DEFAULT_PROTOCOL which is an alias to
@@ -946,7 +978,8 @@ def dumps(obj, protocol=None):
     """
     file = StringIO()
     try:
-        cp = CloudPickler(file, protocol=protocol)
+        cp = CloudPickler(file, protocol=protocol,
+                          override_globals=override_globals)
         cp.dump(obj)
         return file.getvalue()
     finally:
@@ -1078,10 +1111,15 @@ def _fill_function(*args):
     else:
         raise ValueError('Unexpected _fill_value arguments: %r' % (args,))
 
-    # Only set global variables that do not exist.
-    for k, v in state['globals'].items():
-        if k not in func.__globals__:
-            func.__globals__[k] = v
+    # This updates the variables of func's module using func's globals from the
+    # time at which it was pickled.
+    if state.get('override_globals', True):
+        func.__globals__.update(state['globals'])
+    else:
+        for k, v in state['globals'].items():
+            # Only set global variables that do not exist.
+            if k not in func.__globals__:
+                func.__globals__[k] = v
 
     func.__defaults__ = state['defaults']
     func.__dict__ = state['dict']