Merge branch 'release/3.0.2'

* release/3.0.2:
  Removing the changelog. Changelogs are hard.
  Bump to 3.0.2.
  Use == for comparison
  Don't mutate __dict__
  Don't require source in __setstate__; fixes #234
  Add (failing) test for #234
  Don't include cache in serialization of backend
  Fixed name of spec id attr

README.rst

@@ -9,10 +9,9 @@ you. If you need to programatically generate one image from another, you need
 ImageKit.
 
 **For the complete documentation on the latest stable version of ImageKit, see**
-`ImageKit on RTD`_. Our `changelog is also available`_.
+`ImageKit on RTD`_.
 
 .. _`ImageKit on RTD`: http://django-imagekit.readthedocs.org
-.. _`changelog is also available`: http://django-imagekit.readthedocs.org/en/latest/changelog.html
 
 
 Installation

docs/advanced_usage.rst

@@ -47,7 +47,7 @@ for creating an ``ImageSpec``, registering it, and associating it with an
     class Profile(models.Model):
         avatar = models.ImageField(upload_to='avatars')
         avatar_thumbnail = ImageSpecField(source='avatar',
-                                          spec_id='myapp:profile:avatar_thumbnail')
+                                          id='myapp:profile:avatar_thumbnail')
 
 Obviously, the shorthand version is a lot, well…shorter. So why would you ever
 want to go through the trouble of using the long form? The answer is that the
@@ -97,7 +97,7 @@ for getting this information.
     class Profile(models.Model):
         avatar = models.ImageField(upload_to='avatars')
         avatar_thumbnail = ImageSpecField(source='avatar',
-                                          spec_id='myapp:profile:avatar_thumbnail')
+                                          id='myapp:profile:avatar_thumbnail')
         thumbnail_width = models.PositiveIntegerField()
         thumbnail_height = models.PositiveIntegerField()
 

docs/changelog.rst

@@ -1,142 +0,0 @@
-Changelog
-=========
-
-v2.0.2
-------
-
-- Fixed the pickling of ImageSpecFieldFile.
-- Signals are now connected without specifying the class and non-IK models
-  are filitered out in the receivers. This is necessary beacuse of a bug
-  with how Django handles abstract models.
-- Fixed a `ZeroDivisionError` in the Reflection processor.
-- `cStringIO` is now used if it's available.
-- Reflections on images now use RGBA instead of RGB.
-
-v2.0.1
-------
-
-- Fixed a file descriptor leak in the `utils.quiet()` context manager.
-
-
-v2.0.0
-------
-
-- Added the concept of image cache backends. Image cache backends assume
-  control of validating and invalidating the cached images from `ImageSpec` in
-  versions past. The default backend maintins the current behavior: invalidating
-  an image deletes it, while validating checks whether the file exists and
-  creates the file if it doesn't. One can create custom image cache backends to
-  control how their images are cached (e.g., Celery, etc.).
-
-  ImageKit ships with three built-in backends:
-
-  - ``imagekit.imagecache.PessimisticImageCacheBackend`` - A very safe image
-    cache backend. Guarantees that files will always be available, but at the
-    cost of hitting the storage backend.
-  - ``imagekit.imagecache.NonValidatingImageCacheBackend`` - A backend that is
-    super optimistic about the existence of spec files. It will hit your file
-    storage much less frequently than the pessimistic backend, but it is
-    technically possible for a cache file to be missing after validation.
-  - ``imagekit.imagecache.celery.CeleryImageCacheBackend`` - A pessimistic cache
-    state backend that uses celery to generate its spec images. Like
-    ``PessimisticCacheStateBackend``, this one checks to see if the file
-    exists on validation, so the storage is hit fairly frequently, but an
-    image is guaranteed to exist. However, while validation guarantees the
-    existence of *an* image, it does not necessarily guarantee that you will
-    get the correct image, as the spec may be pending regeneration. In other
-    words, while there are ``generate`` tasks in the queue, it is possible to
-    get a stale spec image. The tradeoff is that calling ``invalidate()``
-    won't block to interact with file storage.
-
-- Some of the processors have been renamed and several new ones have been added:
-
-  - ``imagekit.processors.ResizeToFill`` - (previously
-    ``imagekit.processors.resize.Crop``) Scales the image to fill the provided
-    dimensions and then trims away the excess.
-  - ``imagekit.processors.ResizeToFit`` - (previously
-    ``imagekit.processors.resize.Fit``) Scale to fit the provided dimensions.
-  - ``imagekit.processors.SmartResize`` - Like ``ResizeToFill``, but crops using
-    entroy (``SmartCrop``) instead of an anchor argument.
-  - ``imagekit.processors.BasicCrop`` - Crop using provided box.
-  - ``imagekit.processors.SmartCrop`` - (previously
-    ``imagekit.processors.resize.SmartCrop``) Crop to provided size, trimming
-    based on entropy.
-  - ``imagekit.processors.TrimBorderColor`` - Trim the specified color from the
-    specified sides.
-  - ``imagekit.processors.AddBorder`` - Add a border of specific color and
-    thickness to an image.
-  - ``imagekit.processors.Resize`` - Scale to the provided dimensions (can distort).
-  - ``imagekit.processors.ResizeToCover`` - Scale to the smallest size that will
-    cover the specified dimensions. Used internally by ``Fill`` and
-    ``SmartFill``.
-  - ``imagekit.processors.ResizeCanvas`` - Takes an image an resizes the canvas,
-    using a specific background color if the new size is larger than the current
-    image.
-
-- ``mat_color`` has been added as an arguemnt to ``ResizeToFit``. If set, the
-  the target image size will be enforced and the specified color will be
-  used as background color to pad the image.
-
-- We now use `Tox`_ to automate testing.
-
-.. _`Tox`: http://pypi.python.org/pypi/tox
-
-
-v1.1.0
-------
-
-- A ``SmartCrop`` resize processor was added. This allows an image to be
-  cropped based on the amount of entropy in the target image's histogram.
-
-- The ``quality`` argument was removed in favor of an ``options`` dictionary.
-  This is a more general solution which grants access to PIL's format-specific
-  options (including "quality", "progressive", and "optimize" for JPEGs).
-
-- The ``TrimColor`` processor was renamed to ``TrimBorderColor``.
-
-- The private ``_Resize`` class has been removed.
-
-
-v1.0.3
-------
-
-- ``ImageSpec._create()`` was renamed ``ImageSpec.generate()`` and is now
-  available in the public API.
-
-- Added an ``AutoConvert`` processor to encapsulate the transparency
-  handling logic.
-
-- Refactored transparency handling to be smarter, handling a lot more of
-  the situations in which one would convert to or from formats that support
-  transparency.
-
-- Fixed PIL zeroing out files when write mode is enabled.
-
-
-v1.0.2
-------
-
-- Added this changelog.
-
-- Enhanced extension detection, format detection, and conversion between the
-  two. This eliminates the reliance on an image being loaded into memory
-  beforehand in order to detect said image's extension.
-
-- Fixed a regression from the 0.4.x series in which ImageKit was unable to
-  convert a PNG file in ``P`` or "palette" mode to JPEG.
-
-
-v1.0.1
-------
-
-- Minor fixes related to the rendering of ``README.rst`` as a reStructured
-  text file.
-
-- Fixed the included admin template not being found when ImageKit was  and
-  the packaging of the included admin templates.
-
-
-v1.0
-----
-
-- Initial release of the *new* field-based ImageKit API.

docs/index.rst

@@ -21,5 +21,4 @@ Indices and tables
     configuration
     advanced_usage
     caching
-    changelog
     upgrading

imagekit/cachefiles/backends.py

@@ -1,4 +1,5 @@
 from ..utils import get_singleton, sanitize_cache_key
+from copy import copy
 from django.core.cache import get_cache
 from django.core.exceptions import ImproperlyConfigured
 
@@ -71,19 +72,26 @@ class CachedFileBackend(object):
 
     def set_state(self, file, state):
         key = self.get_key(file)
-        if state is CacheFileState.DOES_NOT_EXIST:
+        if state == CacheFileState.DOES_NOT_EXIST:
             self.cache.set(key, state, self.existence_check_timeout)
         else:
             self.cache.set(key, state)
 
+    def __getstate__(self):
+        state = copy(self.__dict__)
+        # Don't include the cache when pickling. It'll be reconstituted based
+        # on the settings.
+        state.pop('_cache', None)
+        return state
+
     def exists(self, file):
-        return self.get_state(file) is CacheFileState.EXISTS
+        return self.get_state(file) == CacheFileState.EXISTS
 
     def generate(self, file, force=False):
         raise NotImplementedError
 
     def generate_now(self, file, force=False):
-        if force or self.get_state(file) is CacheFileState.DOES_NOT_EXIST:
+        if force or self.get_state(file) == CacheFileState.DOES_NOT_EXIST:
             file._generate()
             self.set_state(file, CacheFileState.EXISTS)
 

imagekit/pkgmeta.py

@@ -1,5 +1,5 @@
 __title__ = 'django-imagekit'
 __author__ = 'Justin Driscoll, Bryan Veloso, Greg Newman, Chris Drackett, Matthew Tretter, Eric Eldredge'
-__version__ = '3.0.1'
+__version__ = '3.0.2'
 __license__ = 'BSD'
 __all__ = ['__title__', '__author__', '__version__', '__license__']

imagekit/specs/__init__.py

@@ -1,3 +1,4 @@
+from copy import copy
 from django.conf import settings
 from django.db.models.fields.files import ImageFieldFile
 from ..cachefiles.backends import get_default_cachefile_backend
@@ -93,25 +94,41 @@ class ImageSpec(BaseImageSpec):
         fn = get_by_qname(settings.IMAGEKIT_SPEC_CACHEFILE_NAMER, 'namer')
         return fn(self)
 
+    @property
+    def source(self):
+        src = getattr(self, '_source', None)
+        if not src:
+            field_data = getattr(self, '_field_data', None)
+            if field_data:
+                src = self._source = getattr(field_data['instance'], field_data['attname'])
+                del self._field_data
+        return src
+
+    @source.setter
+    def source(self, value):
+        self._source = value
+
     def __getstate__(self):
-        state = self.__dict__
+        state = copy(self.__dict__)
 
         # Unpickled ImageFieldFiles won't work (they're missing a storage
         # object). Since they're such a common use case, we special case them.
+        # Unfortunately, this also requires us to add the source getter to
+        # lazily retrieve the source on the reconstructed object; simply trying
+        # to look up the source in ``__setstate__`` would require us to get the
+        # model instance but, if ``__setstate__`` was called as part of
+        # deserializing that model, the model wouldn't be fully reconstructed
+        # yet, preventing us from accessing the source field.
+        # (This is issue #234.)
         if isinstance(self.source, ImageFieldFile):
             field = getattr(self.source, 'field')
             state['_field_data'] = {
                 'instance': getattr(self.source, 'instance', None),
                 'attname': getattr(field, 'name', None),
             }
+            state.pop('_source', None)
         return state
 
-    def __setstate__(self, state):
-        field_data = state.pop('_field_data', None)
-        self.__dict__ = state
-        if field_data:
-            self.source = getattr(field_data['instance'], field_data['attname'])
-
     def get_hash(self):
         return hashers.pickle([
             self.source.name,

tests/test_serialization.py

@@ -11,3 +11,15 @@ def test_imagespecfield():
     instance = create_photo('pickletest2.jpg')
     thumbnail = pickleback(instance.thumbnail)
     thumbnail.generate()
+
+
+def test_circular_ref():
+    """
+    A model instance with a spec field in its dict shouldn't raise a KeyError.
+
+    This corresponds to #234
+
+    """
+    instance = create_photo('pickletest3.jpg')
+    instance.thumbnail  # Cause thumbnail to be added to instance's __dict__
+    pickleback(instance)