A first test using PickledObjectField and a rewritten dbserialize module to store Attributes. No migrations set up yet.

src/utils/dbserialize.py

@@ -0,0 +1,282 @@
+"""
+This module handles serialization of arbitrary python structural data,
+intended primarily to be stored in the database. It also supports
+storing Django model instances (which plain pickle cannot do).
+
+This serialization is used internally by the server, notably for
+storing data in Attributes and for piping data to process pools.
+
+The purpose of dbserialize is to handle all forms of data. For
+well-structured non-arbitrary exchange, such as communicating with a
+rich web client, a simpler JSON serialization makes more sense.
+
+This module also implements the SaverList, SaverDict and SaverSet
+classes. These are iterables that track their position in a nested
+structure and makes sure to send updates up to their root. This is
+used by Attributes - without it, one would not be able to update mutables
+in-situ, e.g obj.db.mynestedlist[3][5] = 3 would never be saved and
+be out of sync with the database.
+
+"""
+
+from collections import defaultdict, MutableSequence, MutableSet, MutableMapping
+try:
+    from cPickle import dumps, loads
+except ImportError:
+    from pickle import dumps, loads
+from django.db import transaction
+from django.core.exceptions import ObjectDoesNotExist
+from django.contrib.contenttypes.models import ContentType
+from src.utils.utils import to_str
+
+HIGHEST_PROTOCOL = 2
+
+# initialization and helpers
+
+_GA = object.__getattribute__
+_SA = object.__setattr__
+_FROM_MODEL_MAP = defaultdict(str)
+_FROM_MODEL_MAP.update(dict((c.model, c.natural_key()) for c in ContentType.objects.all()))
+_TO_MODEL_MAP = defaultdict(str)
+_TO_MODEL_MAP.update(dict((c.natural_key(), c.model_class()) for c in ContentType.objects.all()))
+_TO_TYPECLASS = lambda o: hasattr(o, 'typeclass') and o.typeclass or o
+_IS_PACKED_DBOBJ = lambda o: type(o) == tuple and len(o) == 4 and o[0] == '__packed_dbobj__'
+
+
+#
+# SaverList, SaverDict, SaverSet - Attribute-specific helper classes and functions
+#
+
+def _save(method):
+    "method decorator that saves data to Attribute"
+    def save_wrapper(self, *args, **kwargs):
+        ret = method(self, *args, **kwargs)
+        self._save_tree()
+        return ret
+    return save_wrapper
+
+class SaverMutable(object):
+    """
+    Parent class for properly handling  of nested mutables in
+    an Attribute. If not used something like
+     obj.db.mylist[1][2] = "test" (allocation to a nested list)
+    will not save the updated value to the database.
+    """
+    def __init__(self, *args, **kwargs):
+        "store all properties for tracking the tree"
+        self._db_obj = kwargs.pop("db_obj", None)
+        self._parent = None
+        self._data = None
+    def _save_tree(self):
+        "recursively traverse back up the tree, save when we reach the root"
+        if self._parent:
+            self._parent._save_tree()
+        else:
+            try:
+                self._db_obj.value = self
+            except AttributeError:
+                raise AttributeError("SaverMutable %s lacks dobj at its root." % self)
+    def _convert_mutables(self, item):
+        "converts mutables to Saver* variants and assigns .parent property"
+        dtype = type(item)
+        if dtype in (basestring, int, long, float, bool, tuple):
+            return item
+        elif dtype == list:
+            item = SaverList(item)
+            item._parent = self
+        elif dtype == dict:
+            item = SaverDict(item)
+            item._parent = self
+        elif dtype == set:
+            item = SaverSet(item)
+            item._parent = self
+        return item
+    def __repr__(self):
+        return self._data.__repr__()
+    def __len__(self):
+        return self._data.__len__()
+    def __iter__(self):
+        return self._data.__iter__()
+    def __getitem__(self, key):
+        return self._data.__getitem__(key)
+    @_save
+    def __setitem__(self, key, value):
+        self._data.__setitem__(key, self._convert_mutables(value))
+    @_save
+    def __delitem__(self, key):
+        self._data.__delitem__(key)
+
+class SaverList(SaverMutable, MutableSequence):
+    """
+    A list that saves itself to an Attribute when updated.
+    """
+    def __init__(self, *args, **kwargs):
+        super(SaverList, self).__init__(*args, **kwargs)
+        self._data = list(*args)
+    @_save
+    def insert(self, index, value):
+        self._data.insert(index, self._convert_mutables(value))
+
+class SaverDict(SaverMutable, MutableMapping):
+    """
+    A dict that stores changes to an Attribute when updated
+    """
+    def __init__(self, *args, **kwargs):
+        super(SaverDict, self).__init__(*args, **kwargs)
+        self._data = dict(*args)
+
+class SaverSet(SaverMutable, MutableSet):
+    """
+    A set that saves to an Attribute when updated
+    """
+    def __init__(self, *args, **kwargs):
+        super(SaverSet, self).__init__(*args, **kwargs)
+        self._data = set(*args)
+    def __contains__(self, value):
+        return self._data.__contains__(value)
+    @_save
+    def add(self, value):
+        self._data.add(self._convert_mutables(value))
+    @_save
+    def discard(self, value):
+        self._data.discard(value)
+
+
+#
+# serialization access functions
+#
+
+def _pack_dbobj(item):
+    """
+    Check and convert django database objects to an internal representation.
+    This either returns the original input item or a tuple ("__packed_dbobj__", key, obj, id)
+    """
+    obj =  hasattr(item, 'dbobj') and item.dbobj or item
+    natural_key = _FROM_MODEL_MAP[hasattr(obj, "id") and hasattr("db_date_created") and
+                                  hasattr(obj, '__class__') and obj.__class__.__name__.lower()]
+    # build the internal representation as a tuple ("__packed_dbobj__", key, obj, id)
+    return natural_key and ('__packed_dbobj__', natural_key, _GA(obj, "db_date_created"), _GA(obj, id)) or item
+
+def _unpack_dbobj(item):
+    """
+    Check and convert internal representations back to Django database models.
+    The fact that item is a packed dbobj should be checked before this call.
+    This either returns the original input or converts the internal store back
+    to a database representation (its typeclass is returned if applicable).
+    """
+    try:
+        obj = item[3] and _TO_TYPECLASS(_TO_MODEL_MAP[item[1]].objects.get(id=item[3]))
+    except ObjectDoesNotExist:
+        return None
+    # even if we got back a match, check the sanity of the date (some databases may 're-use' the id)
+    return obj and obj.db_data_created == item[3] and obj or None
+
+def to_pickle(data):
+    """
+    This prepares data on arbitrary form to be pickled. It handles any nested structure
+    and returns data on a form that is safe to pickle (including having converted any
+    database models to their internal representation). We also convert any Saver*-type
+    objects back to their normal representations, they are not pickle-safe.
+    """
+    def process_item(item):
+        "Recursive processor and identification of data"
+        dtype = type(item)
+        if dtype in (basestring, int, long, float, bool):
+            return item
+        elif dtype == tuple:
+            return tuple(process_item(val) for val in item)
+        elif dtype in (list, SaverList):
+            return [key for key in item]
+        elif dtype in (dict, SaverDict):
+            return dict((key, process_item(val)) for key, val in item.items())
+        elif dtype in (set, SaverSet):
+            return set(process_item(val) for val in item)
+        elif hasattr(item, '__item__'):
+            # we try to conserve the iterable class, if not convert to list
+            try:
+                return item.__class__([process_item(val) for val in item])
+            except (AttributeError, TypeError):
+                return [process_item(val) for val in item]
+        return _pack_dbobj(item)
+    return process_item(data)
+
+@transaction.autocommit
+def from_pickle(data, db_obj=None):
+    """
+    This should be fed a just de-pickled data object. It will be converted back
+    to a form that may contain database objects again. Note that if a database
+    object was removed (or changed in-place) in the database, None will be returned.
+
+    db_obj - this is the model instance (normally an Attribute) that Saver*-type
+             iterables will save to when they update. It must have a 'value'
+             property that saves assigned data to the database.
+
+    If db_obj is given, this function will convert lists, dicts and sets to their
+    SaverList, SaverDict and SaverSet counterparts.
+
+    """
+    def process_item(item):
+        "Recursive processor and identification of data"
+        dtype = type(item)
+        if dtype in (basestring, int, long, float, bool):
+            return item
+        elif _IS_PACKED_DBOBJ(item):
+            # this must be checked before tuple
+            return _unpack_dbobj(item)
+        elif dtype == tuple:
+            return tuple(process_item(val) for val in item)
+        elif dtype == dict:
+            return dict((key, process_item(val)) for key, val in item.items())
+        elif dtype == set:
+            return set(process_item(val) for val in item)
+        elif hasattr(item, '__iter__'):
+            try:
+                # we try to conserve the iterable class if it accepts an iterator
+                return item.__class__(process_item(val) for val in item)
+            except (AttributeError, TypeError):
+                return [process_item(val) for val in item]
+        return item
+
+    def process_item_to_savers(item):
+        "Recursive processor, convertion and identification of data"
+        dtype = type(item)
+        if dtype in (basestring, int, long, float, bool):
+            return item
+        elif _IS_PACKED_DBOBJ(item):
+            # this must be checked before tuple
+            return _unpack_dbobj(item)
+        elif dtype == tuple:
+            return tuple(process_item_to_savers(val) for val in item)
+        elif dtype == list:
+            return SaverList(process_item_to_savers(val) for val in item)
+        elif dtype == dict:
+            return SaverDict((key, process_item_to_savers(val)) for key, val in item.items())
+        elif dtype == set:
+            return SaverSet(process_item_to_savers(val) for val in item)
+        elif hasattr(item, '__iter__'):
+            try:
+                # we try to conserve the iterable class if it accepts an iterator
+                return item.__class__(process_item_to_savers(val) for val in item)
+            except (AttributeError, TypeError):
+                return SaverList(process_item_to_savers(val) for val in item)
+        return item
+
+    if db_obj:
+        # convert lists, dicts and sets to their Saved* counterparts. It
+        # is only relevant if the "root" is an iterable of the right type.
+        dtype = type(data)
+        if dtype == list:
+            return process_item_to_savers(SaverList(data, db_obj=db_obj))
+        elif dtype == dict:
+            return process_item_to_savers(SaverDict(data, db_obj=db_obj))
+        elif dtype == set:
+            return process_item_to_savers(SaverSet(data, db_obj=db_obj))
+    return process_item(data)
+
+def do_pickle(data):
+    "Perform pickle to string"
+    return to_str(dumps(data, protocol=HIGHEST_PROTOCOL))
+
+def do_unpickle(data):
+    "Retrieve pickle from pickled string"
+    return loads(to_str(data))