Added indices to all Attribute fields to improve lookup speed. Added related_lookup_f...
[philo.git] / models / base.py
index ae15d16..5f8fb03 100644 (file)
@@ -2,14 +2,17 @@ from django import forms
 from django.db import models
 from django.contrib.contenttypes.models import ContentType
 from django.contrib.contenttypes import generic
-from django.utils import simplejson as json
 from django.core.exceptions import ObjectDoesNotExist
+from django.core.validators import RegexValidator
+from django.utils import simplejson as json
+from django.utils.encoding import smart_str
 from philo.exceptions import AncestorDoesNotExist
 from philo.models.fields import JSONField
 from philo.utils import ContentTypeRegistryLimiter, ContentTypeSubclassLimiter
 from philo.signals import entity_class_prepared
 from philo.validators import json_validator
 from UserDict import DictMixin
+from mptt.models import MPTTModel, MPTTModelBase, MPTTOptions
 
 
 class Tag(models.Model):
@@ -21,6 +24,7 @@ class Tag(models.Model):
        
        class Meta:
                app_label = 'philo'
+               ordering = ('name',)
 
 
 class Titled(models.Model):
@@ -41,6 +45,9 @@ def register_value_model(model):
        value_content_type_limiter.register_class(model)
 
 
+register_value_model(Tag)
+
+
 def unregister_value_model(model):
        value_content_type_limiter.unregister_class(model)
 
@@ -52,10 +59,15 @@ class AttributeValue(models.Model):
        def attribute(self):
                return self.attribute_set.all()[0]
        
-       def apply_data(self, data):
+       def set_value(self, value):
                raise NotImplementedError
        
-       def value_formfield(self, **kwargs):
+       def value_formfields(self, **kwargs):
+               """Define any formfields that would be used to construct an instance of this value."""
+               raise NotImplementedError
+       
+       def construct_instance(self, **kwargs):
+               """Apply cleaned data from the formfields generated by valid_formfields to oneself."""
                raise NotImplementedError
        
        def __unicode__(self):
@@ -69,17 +81,22 @@ attribute_value_limiter = ContentTypeSubclassLimiter(AttributeValue)
 
 
 class JSONValue(AttributeValue):
-       value = JSONField() #verbose_name='Value (JSON)', help_text='This value must be valid JSON.')
+       value = JSONField(verbose_name='Value (JSON)', help_text='This value must be valid JSON.', default='null')
        
        def __unicode__(self):
-               return self.value_json
+               return smart_str(self.value)
        
-       def value_formfield(self, **kwargs):
-               kwargs['initial'] = self.value_json
-               return self._meta.get_field('value').formfield(**kwargs)
+       def value_formfields(self):
+               kwargs = {'initial': self.value_json}
+               field = self._meta.get_field('value')
+               return {field.name: field.formfield(**kwargs)}
        
-       def apply_data(self, cleaned_data):
-               self.value = cleaned_data.get('value', None)
+       def construct_instance(self, **kwargs):
+               field_name = self._meta.get_field('value').name
+               self.set_value(kwargs.pop(field_name, None))
+       
+       def set_value(self, value):
+               self.value = value
        
        class Meta:
                app_label = 'philo'
@@ -90,19 +107,33 @@ class ForeignKeyValue(AttributeValue):
        object_id = models.PositiveIntegerField(verbose_name='Value ID', null=True, blank=True)
        value = generic.GenericForeignKey()
        
-       def value_formfield(self, form_class=forms.ModelChoiceField, **kwargs):
-               if self.content_type is None:
-                       return None
-               kwargs.update({'initial': self.object_id, 'required': False})
-               return form_class(self.content_type.model_class()._default_manager.all(), **kwargs)
-       
-       def apply_data(self, cleaned_data):
-               if 'value' in cleaned_data and cleaned_data['value'] is not None:
-                       self.value = cleaned_data['value']
-               else:
-                       self.content_type = cleaned_data.get('content_type', None)
-                       # If there is no value set in the cleaned data, clear the stored value.
+       def value_formfields(self):
+               field = self._meta.get_field('content_type')
+               fields = {field.name: field.formfield(initial=getattr(self.content_type, 'pk', None))}
+               
+               if self.content_type:
+                       kwargs = {
+                               'initial': self.object_id,
+                               'required': False,
+                               'queryset': self.content_type.model_class()._default_manager.all()
+                       }
+                       fields['value'] = forms.ModelChoiceField(**kwargs)
+               return fields
+       
+       def construct_instance(self, **kwargs):
+               field_name = self._meta.get_field('content_type').name
+               ct = kwargs.pop(field_name, None)
+               if ct is None or ct != self.content_type:
                        self.object_id = None
+                       self.content_type = ct
+               else:
+                       value = kwargs.pop('value', None)
+                       self.set_value(value)
+                       if value is None:
+                               self.content_type = ct
+       
+       def set_value(self, value):
+               self.value = value
        
        class Meta:
                app_label = 'philo'
@@ -112,72 +143,91 @@ class ManyToManyValue(AttributeValue):
        content_type = models.ForeignKey(ContentType, limit_choices_to=value_content_type_limiter, verbose_name='Value type', null=True, blank=True)
        values = models.ManyToManyField(ForeignKeyValue, blank=True, null=True)
        
-       def get_object_id_list(self):
-               if not self.values.count():
-                       return []
-               else:
-                       return self.values.values_list('object_id', flat=True)
-       
-       def get_value(self):
-               if self.content_type is None:
-                       return None
-               
-               return self.content_type.model_class()._default_manager.filter(id__in=self.get_object_id_list())
+       def get_object_ids(self):
+               return self.values.values_list('object_id', flat=True)
+       object_ids = property(get_object_ids)
        
        def set_value(self, value):
-               # Value is probably a queryset - but allow any iterable.
+               # Value must be a queryset. Watch out for ModelMultipleChoiceField;
+               # it returns its value as a list if empty.
                
-               # These lines shouldn't be necessary; however, if value is an EmptyQuerySet,
-               # the code (specifically the object_id__in query) won't work without them. Unclear why...
-               if not value:
-                       value = []
+               self.content_type = ContentType.objects.get_for_model(value.model)
                
                # Before we can fiddle with the many-to-many to foreignkeyvalues, we need
                # a pk.
                if self.pk is None:
                        self.save()
                
-               if isinstance(value, models.query.QuerySet):
-                       value = value.values_list('id', flat=True)
-               
-               self.values.filter(~models.Q(object_id__in=value)).delete()
-               current = self.get_object_id_list()
+               object_ids = value.values_list('id', flat=True)
                
-               for v in value:
-                       if v in current:
-                               continue
-                       self.values.create(content_type=self.content_type, object_id=v)
-       
-       value = property(get_value, set_value)
+               # These lines shouldn't be necessary; however, if object_ids is an EmptyQuerySet,
+               # the code (specifically the object_id__in query) won't work without them. Unclear why...
+               # TODO: is this still the case?
+               if not object_ids:
+                       self.values.all().delete()
+               else:
+                       self.values.exclude(object_id__in=object_ids, content_type=self.content_type).delete()
+                       
+                       current_ids = self.object_ids
+                       
+                       for object_id in object_ids:
+                               if object_id in current_ids:
+                                       continue
+                               self.values.create(content_type=self.content_type, object_id=object_id)
        
-       def value_formfield(self, form_class=forms.ModelMultipleChoiceField, **kwargs):
+       def get_value(self):
                if self.content_type is None:
                        return None
-               kwargs.update({'initial': self.get_object_id_list(), 'required': False})
-               return form_class(self.content_type.model_class()._default_manager.all(), **kwargs)
+               
+               # HACK to be safely explicit until http://code.djangoproject.com/ticket/15145 is resolved
+               object_ids = self.object_ids
+               manager = self.content_type.model_class()._default_manager
+               if not object_ids:
+                       return manager.none()
+               return manager.filter(id__in=self.object_ids)
        
-       def apply_data(self, cleaned_data):
-               if 'value' in cleaned_data and cleaned_data['value'] is not None:
-                       self.value = cleaned_data['value']
+       value = property(get_value, set_value)
+       
+       def value_formfields(self):
+               field = self._meta.get_field('content_type')
+               fields = {field.name: field.formfield(initial=getattr(self.content_type, 'pk', None))}
+               
+               if self.content_type:
+                       kwargs = {
+                               'initial': self.object_ids,
+                               'required': False,
+                               'queryset': self.content_type.model_class()._default_manager.all()
+                       }
+                       fields['value'] = forms.ModelMultipleChoiceField(**kwargs)
+               return fields
+       
+       def construct_instance(self, **kwargs):
+               field_name = self._meta.get_field('content_type').name
+               ct = kwargs.pop(field_name, None)
+               if ct is None or ct != self.content_type:
+                       self.values.clear()
+                       self.content_type = ct
                else:
-                       self.content_type = cleaned_data.get('content_type', None)
-                       # If there is no value set in the cleaned data, clear the stored value.
-                       self.value = []
+                       value = kwargs.get('value', None)
+                       if not value:
+                               value = self.content_type.model_class()._default_manager.none()
+                       self.set_value(value)
+       construct_instance.alters_data = True
        
        class Meta:
                app_label = 'philo'
 
 
 class Attribute(models.Model):
-       entity_content_type = models.ForeignKey(ContentType, related_name='attribute_entity_set', verbose_name='Entity type')
-       entity_object_id = models.PositiveIntegerField(verbose_name='Entity ID')
+       entity_content_type = models.ForeignKey(ContentType, related_name='attribute_entity_set', verbose_name='Entity type', db_index=True)
+       entity_object_id = models.PositiveIntegerField(verbose_name='Entity ID', db_index=True)
        entity = generic.GenericForeignKey('entity_content_type', 'entity_object_id')
        
-       value_content_type = models.ForeignKey(ContentType, related_name='attribute_value_set', limit_choices_to=attribute_value_limiter, verbose_name='Value type', null=True, blank=True)
-       value_object_id = models.PositiveIntegerField(verbose_name='Value ID', null=True, blank=True)
+       value_content_type = models.ForeignKey(ContentType, related_name='attribute_value_set', limit_choices_to=attribute_value_limiter, verbose_name='Value type', null=True, blank=True, db_index=True)
+       value_object_id = models.PositiveIntegerField(verbose_name='Value ID', null=True, blank=True, db_index=True)
        value = generic.GenericForeignKey('value_content_type', 'value_object_id')
        
-       key = models.CharField(max_length=255)
+       key = models.CharField(max_length=255, validators=[RegexValidator("\w+")], help_text="Must contain one or more alphanumeric characters or underscores.", db_index=True)
        
        def __unicode__(self):
                return u'"%s": %s' % (self.key, self.value)
@@ -241,37 +291,6 @@ class Entity(models.Model):
        def attributes(self):
                return QuerySetMapper(self.attribute_set.all())
        
-       @property
-       def _added_attribute_registry(self):
-               if not hasattr(self, '_real_added_attribute_registry'):
-                       self._real_added_attribute_registry = {}
-               return self._real_added_attribute_registry
-       
-       @property
-       def _removed_attribute_registry(self):
-               if not hasattr(self, '_real_removed_attribute_registry'):
-                       self._real_removed_attribute_registry = []
-               return self._real_removed_attribute_registry
-       
-       def save(self, *args, **kwargs):
-               super(Entity, self).save(*args, **kwargs)
-               
-               for key in self._removed_attribute_registry:
-                       self.attribute_set.filter(key__exact=key).delete()
-               del self._removed_attribute_registry[:]
-               
-               for field, value in self._added_attribute_registry.items():
-                       try:
-                               attribute = self.attribute_set.get(key__exact=field.key)
-                       except Attribute.DoesNotExist:
-                               attribute = Attribute()
-                               attribute.entity = self
-                               attribute.key = field.key
-                       
-                       field.set_attribute_value(attribute, value)
-                       attribute.save()
-               self._added_attribute_registry.clear()
-       
        class Meta:
                abstract = True
 
@@ -279,23 +298,23 @@ class Entity(models.Model):
 class TreeManager(models.Manager):
        use_for_related_fields = True
        
-       def roots(self):
-               return self.filter(parent__isnull=True)
-       
        def get_with_path(self, path, root=None, absolute_result=True, pathsep='/', field='slug'):
                """
                Returns the object with the path, unless absolute_result is set to False, in which
                case it returns a tuple containing the deepest object found along the path, and the
                remainder of the path after that object as a string (or None if there is no remaining
                path). Raises a DoesNotExist exception if no object is found with the given path.
+               
+               If the path you're searching for is known to exist, it is always faster to use
+               absolute_result=True - unless the path depth is over ~40, in which case the high cost
+               of the absolute query makes a binary search (i.e. non-absolute) faster.
                """
+               # Note: SQLite allows max of 64 tables in one join. That means the binary search will
+               # only work on paths with a max depth of 127 and the absolute fetch will only work
+               # to a max depth of (surprise!) 63. Although this could be handled, chances are your
+               # tree structure won't be that deep.
                segments = path.split(pathsep)
                
-               # Check for a trailing pathsep so we can restore it later.
-               trailing_pathsep = False
-               if segments[-1] == '':
-                       trailing_pathsep = True
-               
                # Clean out blank segments. Handles multiple consecutive pathseps.
                while True:
                        try:
@@ -306,11 +325,13 @@ class TreeManager(models.Manager):
                # Special-case a lack of segments. No queries necessary.
                if not segments:
                        if root is not None:
+                               if absolute_result:
+                                       return root
                                return root, None
                        else:
                                raise self.model.DoesNotExist('%s matching query does not exist.' % self.model._meta.object_name)
                
-               def make_query_kwargs(segments):
+               def make_query_kwargs(segments, root):
                        kwargs = {}
                        prefix = ""
                        revsegs = list(segments)
@@ -320,72 +341,88 @@ class TreeManager(models.Manager):
                                kwargs["%s%s__exact" % (prefix, field)] = segment
                                prefix += "parent__"
                        
-                       kwargs[prefix[:-2]] = root
+                       if prefix:
+                               kwargs[prefix[:-2]] = root
+                       
                        return kwargs
                
-               def build_path(segments):
-                       path = pathsep.join(segments)
-                       if trailing_pathsep and segments and segments[-1] != '':
-                               path += pathsep
-                       return path
-               
-               def find_obj(segments, depth, deepest_found):
+               def find_obj(segments, depth, deepest_found=None):
+                       if deepest_found is None:
+                               deepest_level = 0
+                       elif root is None:
+                               deepest_level = deepest_found.get_level() + 1
+                       else:
+                               deepest_level = deepest_found.get_level() - root.get_level()
                        try:
-                               obj = self.get(**make_query_kwargs(segments[:depth]))
+                               obj = self.get(**make_query_kwargs(segments[deepest_level:depth], deepest_found or root))
                        except self.model.DoesNotExist:
-                               if absolute_result:
-                                       raise
+                               if not deepest_level and depth > 1:
+                                       # make sure there's a root node...
+                                       depth = 1
+                               else:
+                                       # Try finding one with half the path since the deepest find.
+                                       depth = (deepest_level + depth)/2
                                
-                               depth = (deepest_found + depth)/2
-                               if deepest_found == depth:
+                               if deepest_level == depth:
                                        # This should happen if nothing is found with any part of the given path.
+                                       if root is not None and deepest_found is None:
+                                               return root, pathsep.join(segments)
                                        raise
                                
-                               # Try finding one with half the path since the deepest find.
                                return find_obj(segments, depth, deepest_found)
                        else:
-                               # Yay! Found one! Could there be a deeper one?
-                               if absolute_result:
-                                       return obj
+                               # Yay! Found one!
+                               if root is None:
+                                       deepest_level = obj.get_level() + 1
+                               else:
+                                       deepest_level = obj.get_level() - root.get_level()
+                               
+                               # Could there be a deeper one?
+                               if obj.is_leaf_node():
+                                       return obj, pathsep.join(segments[deepest_level:]) or None
                                
-                               deepest_found = depth
-                               depth = (len(segments) + depth)/2
+                               depth += (len(segments) - depth)/2 or len(segments) - depth
                                
-                               if deepest_found == depth:
-                                       return obj, build_path(segments[deepest_found:]) or None
+                               if depth > deepest_level + obj.get_descendant_count():
+                                       depth = deepest_level + obj.get_descendant_count()
+                               
+                               if deepest_level == depth:
+                                       return obj, pathsep.join(segments[deepest_level:]) or None
                                
                                try:
-                                       return find_obj(segments, depth, deepest_found)
+                                       return find_obj(segments, depth, obj)
                                except self.model.DoesNotExist:
-                                       # Then the deepest one was already found.
-                                       return obj, build_path(segments[deepest_found:])
+                                       # Then this was the deepest.
+                                       return obj, pathsep.join(segments[deepest_level:])
+               
+               if absolute_result:
+                       return self.get(**make_query_kwargs(segments, root))
                
-               return find_obj(segments, len(segments), 0)
+               # Try a modified binary search algorithm. Feed the root in so that query complexity
+               # can be reduced. It might be possible to weight the search towards the beginning
+               # of the path, since short paths are more likely, but how far forward? It would
+               # need to shift depending on len(segments) - perhaps logarithmically?
+               return find_obj(segments, len(segments)/2 or len(segments))
 
 
-class TreeModel(models.Model):
+class TreeModel(MPTTModel):
        objects = TreeManager()
        parent = models.ForeignKey('self', related_name='children', null=True, blank=True)
        slug = models.SlugField(max_length=255)
        
-       def has_ancestor(self, ancestor):
-               parent = self
-               while parent:
-                       if parent == ancestor:
-                               return True
-                       parent = parent.parent
-               return False
-       
        def get_path(self, root=None, pathsep='/', field='slug'):
-               if root is not None and not self.has_ancestor(root):
+               if root == self:
+                       return ''
+               
+               if root is not None and not self.is_descendant_of(root):
                        raise AncestorDoesNotExist(root)
                
-               path = getattr(self, field, '?')
-               parent = self.parent
-               while parent and parent != root:
-                       path = getattr(parent, field, '?') + pathsep + path
-                       parent = parent.parent
-               return path
+               qs = self.get_ancestors(include_self=True)
+               
+               if root is not None:
+                       qs = qs.filter(**{'%s__gt' % self._mptt_meta.level_attr: root.get_level()})
+               
+               return pathsep.join([getattr(parent, field, '?') for parent in qs])
        path = property(get_path)
        
        def __unicode__(self):
@@ -396,7 +433,17 @@ class TreeModel(models.Model):
                abstract = True
 
 
+class TreeEntityBase(MPTTModelBase, EntityBase):
+       def __new__(meta, name, bases, attrs):
+               attrs['_mptt_meta'] = MPTTOptions(attrs.pop('MPTTMeta', None))
+               cls = EntityBase.__new__(meta, name, bases, attrs)
+               
+               return meta.register(cls)
+
+
 class TreeEntity(Entity, TreeModel):
+       __metaclass__ = TreeEntityBase
+       
        @property
        def attributes(self):
                if self.parent: