Improvement/performance tweaks [WIP]

docs/guide/serializers.rst

@@ -110,25 +110,21 @@ All field classes accept this set of arguments:
   intead of relying on param labels.*
 
 * **source** *(str, optional):* name of internal object key/attribute
-  that will be passed to field's on ``.to_representation(value)`` call.
-  Special ``'*'`` value is allowed that will pass whole object to
-  field when making representation. If not set then default source will
-  be a field name used as a serializer's attribute.
+  that will be passed to field's on ``.to_representation(value)`` call. If not
+  set then default source is a field name used as a serializer's attribute.
 
 * **validators** *(list, optional):* list of validator callables.
 
 * **many** *(bool, optional)* set to True if field is in fact a list
   of given type objects
 
 
-.. note::
-
-   ``source='*'`` is in fact a dirty workaround and will not work well
-   on validation when new object instances needs to be created/updated
-   using POST/PUT requests. This works quite well with simple retrieve/list
-   type resources but in more sophisticated cases it is better to use
-   custom object properties as sources to encapsulate such fields.
+.. versionchanged:: 1.0.0
 
+   Fields no no longer have special case treatment for ``source='*'`` argument.
+   If you want to access multiple object keys and values within single
+   serializer field please refer to :ref:`guide-field-attribute-access` section
+   of this document.
 
 .. _field-validation:
 
@@ -143,9 +139,9 @@ in order to provide correct HTTP responses each validator shoud raise
 .. note::
 
    Concept of validation for fields is understood here as a process of checking
-   if data of valid type (successfully parsed/processed by
-   ``.from_representation`` handler) does meet some other constraints
-   (lenght, bounds, unique, etc).
+   if data of valid type (i.e. data that was successfully parsed/processed by
+   ``.from_representation()`` handler) does meet some other constraints
+   (lenght, bounds, uniquess, etc).
 
 
 Example of simple validator usage:
@@ -174,31 +170,50 @@ Resource validation
 ~~~~~~~~~~~~~~~~~~~
 
 In most cases field level validation is all that you need but sometimes you
-need to perfom obejct level validation that needs to access multiple fields
-that are already deserialized and validated. Suggested way to do this in
-graceful is to override serializer's ``.validate()`` method and raise
-:class:`graceful.errors.ValidationError` when your validation fails. This
-exception will be then automatically translated to HTTP Bad Request response
-on resource-level handlers. Here is example:
+need to perfom validation on whole resource representation or deserialized
+object. It is possible to access multiple fields that were already deserialized
+and pre-validated directly from serializer class.
+
+You can provide your own object-level serialization handler using serializer's
+``validate()`` method. This method accepts two arguments:
+
+* **object_dict** *(dict):* it is deserialized object dictionary that already
+  passed validation. Field sources instead of their representation names are
+  used as its keys.
+
+* **partial** *(bool):* it is set to ``True`` only on partial object updates
+  (e.g. on ``PATCH`` requests). If you plan to support partial resource
+  modification you should check this field and verify if you object has
+  all the existing keys.
+
+If your validation fails you should raise the
+:class:`graceful.errors.ValidationError` exception. Following is the example
+of resource serializer with custom object-level validation:
 
 
 .. code-block:: python
 
     class DrinkSerializer():
-        alcohol = StringField("main ingredient", required=True)
-        mixed_with = StringField("what makes it tasty", required=True)
-
-        def validate(self, object_dict, partial=False):
-            # note: always make sure to call super `validate()`
-            # so whole validation of fields works as expected
-            super().validate(object_dict, partial)
+        alcohol = StringField("main ingredient")
+        mixed_with = StringField("what makes it tasty")
+
+        def validate(self, object_dict, partial):
+            if partial and any([
+                'alcohol' in object_dict,
+                'mixed_with' in object_dict,
+            ]):
+                raise ValidationError(
+                    "bartender refused to change ingredients"
+                )
 
             # here is a place for your own validation
             if (
                 object_dict['alcohol'] == 'whisky' and
                 object_dict['mixed_with'] == 'cola'
             ):
-                raise ValidationError("bartender refused!')
+                raise ValidationError(
+                    "bartender refused to mix whisky with cola!"
+                )
 
 
 Custom fields
@@ -228,3 +243,90 @@ as a serialized JSON string that we would like to (de)serialize:
         def to_representation(data):
             return json.loads(data)
 
+.. _guide-field-attribute-access:
+
+
+Accessing multiple fields at once
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Sometimes you need to access multiple fields of internal object instance at
+once in order to properly represent data in your API. This is very common when
+interacting with legacy services/components that cannot be changed or when
+your storage engine simply does not allow to store nested or structured objects.
+
+Serializers generally work on per-field basis and allow only to translate field
+names between representation and application internal objects. In order to
+manipulate multiple representation or internal object instance keys within the
+single field you need to create custom field class and override one or more
+of following methods:
+
+* ``read_instance(self, instance, source)``: read value from the
+  object instance before serialization. The return value will be later passed
+  as an argument to ``to_representation()`` method. The ``source`` argument is
+  field's name or configured source name (if custom ``source`` is explicitly
+  specified for that field). Base implementation defaults to dictionary key
+  lookup or object attribute lookup.
+
+* ``update_instance(self, instance, source, value)``: update the
+  content of object instance after deserialization. The ``value`` argument is
+  the return value of ``from_representation()`` method. The ``source`` argument
+  is the field's name or source name (if custom ``source`` is explicitly
+  specified for that field). Base implementation defaults to dictionary key
+  assignment or object attribute assignment.
+
+* ``read_representation(self, representation, field_name)``: read value
+  from the representation before deserialization. The return value will be
+  later passed as an argument to ``from_representation()`` method. The
+  ``field_name`` argument is a field's name. Base implementation defaults
+  to dictionary key lookup or object attribute lookup.
+
+* ``update_representation(self, representation, field_name, value)``:
+  update the content of representation instance after serialization.
+  The ``value`` argument is the return value of ``to_representation()`` method.
+  The ``field_name`` argument is a field's name. Base implementation
+  defaults to dictionary key assignment.
+
+To better explain how to use these methods let's assume that due to some
+storage backend constraints we cannot save nested dictionaries. All of fields
+of some nested object will have to be stored under separate keys but we still
+want to present this to the user as separate nested dictionary. And of course
+we want to support both writes and saves.
+
+.. code-block:: python
+
+    class OwnerField(RawField):
+        def from_representation(self, data):
+            if not isinstance(data, dict):
+                raise ValueError("expected object")
+
+            return {
+                'owner_name': data.get('name'),
+                'owner_age': data.get('age'),
+            }
+
+        def to_representation(self, value):
+            return {
+                'age': value.get('owner_age'),
+                'name': value.get('owner_name'),
+            }
+
+        def validate(self, value):
+            if 'owner_age' not in value or not isinstance(value['owner_age'], int):
+                raise ValidationError("invalid owner age")
+
+            if 'owner_name' not in value:
+                raise ValidationError("invalid owner name")
+
+        def update_instance(self, instance, source, value):
+            # we assume that instance is always a dictionary so we can
+            # use the .update() method
+            instance.update(value)
+
+        def read_instance(self, instance, source):
+            # .to_representation() method requires acces to whole object
+            # dictionary so we have to return whole object.
+            return instance
+
+
+Similar approach may be used to flatten nested objects into more compact
+representations.

src/graceful/errors.py

@@ -34,7 +34,7 @@ def _get_description(self):
                     "forbidden: {}".format(self.forbidden)
                     if self.forbidden else ""
                 ),
-                "invalid: {}:".format(self.invalid) if self.invalid else "",
+                "invalid: {}".format(self.invalid) if self.invalid else "",
                 (
                     "failed to parse: {}".format(self.failed)
                     if self.failed else ""

src/graceful/fields.py

@@ -1,4 +1,5 @@
 import inspect
+from collections import Mapping, MutableMapping
 
 from graceful.validators import min_validator, max_validator
 
@@ -60,6 +61,14 @@ def from_representation(self, data):
             def to_representation(self, value):
                 return ["True", "False"][value]
 
+    .. versionchanged:: 1.0.0
+        Field instances no longer support ``source="*"`` to access whole object
+        for the purpose of representation serialization/deserialization. If you
+        want to access multiple fields of object instance and/or its
+        representation you must override ``read_*`` and ``update_*`` methods in
+        your custom field classes (see :ref:`guide-field-attribute-access`
+        section in documentation).
+
     """
 
     #: Two-tuple ``(label, url)`` pointing to represented type specification
@@ -172,6 +181,64 @@ def validate(self, value):
         for validator in self.validators:
             validator(value)
 
+    def update_instance(self, instance, source, value):
+        """Update object instance after deserialization.
+
+        Args:
+            instance (object): dictionary or object after serialization.
+            source (str): field's name or source if ``source`` was
+                explicitly specified for that field.
+            value (object): return value from ``from_representation`` method.
+        """
+        if isinstance(instance, MutableMapping):
+            instance[source] = value
+        else:
+            setattr(instance, source, value)
+
+    def read_instance(self, instance, source):
+        """Read value from the object instance before serialization.
+
+        Args:
+            instance (object): dictionary or object before serialization.
+            source (str): field's name or source if ``source`` was
+                explicitly specified for that field.
+
+        Returns:
+            The value that will be later passed as an argument to
+            ``to_representation()`` method.
+        """
+        if isinstance(instance, Mapping):
+            return instance.get(source, None)
+
+        return getattr(instance, source, None)
+
+    def update_representation(self, representation, field_name, value):
+        """Update representation after field serialization.
+
+        Args:
+            representation (dict): representation dictionary.
+            attribute_or_key (str): field's name.
+            value (object): return value from ``to_representation`` method.
+        """
+        # note: Unlike instances, representations can only be dictionaries
+        #       and their type cannot be overriden inside of serializer's class
+        representation[field_name] = value
+
+    def read_representation(self, representation, field_name):
+        """Read value from the representation before deserialization.
+
+        Args:
+            representation (dict): representation dictionary.
+            field_name (str): field's name.
+
+        Returns:
+            The value that will be later passed as an argument to
+            ``from_representation()`` method.
+        """
+        # note: Unlike instances, representations can only be dictionaries
+        #       and their type cannot be overriden inside of serializer's class
+        return representation.get(field_name, None)
+
 
 class RawField(BaseField):
     """Represents raw field subtype.

src/graceful/resources/base.py

@@ -424,11 +424,11 @@ def require_validated(self, req, partial=False, bulk=False):
 
         try:
             for representation in representations:
-                object_dict = self.serializer.from_representation(
-                    representation
+                object_dicts.append(
+                    self.serializer.from_representation(
+                        representation, partial
+                    )
                 )
-                self.serializer.validate(object_dict, partial)
-                object_dicts.append(object_dict)
 
         except DeserializationError as err:
             # when working on Resource we know that we can finally raise