Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Browse files

Merge branch 'mapreduce' of git://github.com/blackbrrr/mongoengine

Conflicts:
	mongoengine/queryset.py
  • Loading branch information...
commit 047cc218a64a747e948c537bd2fc404d45d214ac 2 parents 39fc862 + f47d926
Harry Marr hmarr authored
3  docs/apireference.rst
View
@@ -20,6 +20,9 @@ Documents
.. autoclass:: mongoengine.EmbeddedDocument
:members:
+
+.. autoclass:: mongoengine.MapReduceDocument
+ :members:
Querying
========
2  docs/conf.py
View
@@ -22,7 +22,7 @@
# Add any Sphinx extension module names here, as strings. They can be extensions
# coming with Sphinx (named 'sphinx.ext.*') or your custom ones.
-extensions = ['sphinx.ext.autodoc']
+extensions = ['sphinx.ext.autodoc', 'sphinx.ext.todo']
# Add any paths that contain templates here, relative to this directory.
templates_path = ['_templates']
36 mongoengine/document.py
View
@@ -115,3 +115,39 @@ def drop_collection(cls):
"""
db = _get_db()
db.drop_collection(cls._meta['collection'])
+
+
+class MapReduceDocument(object):
+ """A document returned from a map/reduce query.
+
+ :param collection: An instance of :class:`~pymongo.Collection`
+ :param key: Document/result key, often an instance of
+ :class:`~pymongo.objectid.ObjectId`. If supplied as
+ an ``ObjectId`` found in the given ``collection``,
+ the object can be accessed via the ``object`` property.
+ :param value: The result(s) for this key.
+
+ .. versionadded:: 0.3
+
+ """
+
+ def __init__(self, document, collection, key, value):
+ self._document = document
+ self._collection = collection
+ self.key = key
+ self.value = value
+
+ @property
+ def object(self):
+ """Lazy-load the object referenced by ``self.key``. If ``self.key``
+ is not an ``ObjectId``, simply return ``self.key``.
+ """
+ if not isinstance(self.key, (pymongo.objectid.ObjectId)):
+ try:
+ self.key = pymongo.objectid.ObjectId(self.key)
+ except:
+ return self.key
+ if not hasattr(self, "_key_object"):
+ self._key_object = self._document.objects.with_id(self.key)
+ return self._key_object
+ return self._key_object
131 mongoengine/queryset.py
View
@@ -5,7 +5,7 @@
import copy
-__all__ = ['queryset_manager', 'Q', 'InvalidQueryError',
+__all__ = ['queryset_manager', 'Q', 'InvalidQueryError',
'InvalidCollectionError']
# The maximum number of items to display in a QuerySet.__repr__
@@ -13,7 +13,7 @@
class DoesNotExist(Exception):
- pass
+ pass
class MultipleObjectsReturned(Exception):
@@ -30,8 +30,9 @@ class OperationError(Exception):
RE_TYPE = type(re.compile(''))
+
class Q(object):
-
+
OR = '||'
AND = '&&'
OPERATORS = {
@@ -52,7 +53,7 @@ class Q(object):
'regex_eq': '%(value)s.test(this.%(field)s)',
'regex_ne': '!%(value)s.test(this.%(field)s)',
}
-
+
def __init__(self, **query):
self.query = [query]
@@ -132,10 +133,10 @@ def _build_op_js(self, op, key, value, value_name):
return value, operation_js
class QuerySet(object):
- """A set of results returned from a query. Wraps a MongoDB cursor,
+ """A set of results returned from a query. Wraps a MongoDB cursor,
providing :class:`~mongoengine.Document` objects as the results.
"""
-
+
def __init__(self, document, collection):
self._document = document
self._collection_obj = collection
@@ -143,7 +144,8 @@ def __init__(self, document, collection):
self._query = {}
self._where_clause = None
self._loaded_fields = []
-
+ self._ordering = []
+
# If inheritance is allowed, only return instances and instances of
# subclasses of the class being used
if document._meta.get('allow_inheritance'):
@@ -151,7 +153,7 @@ def __init__(self, document, collection):
self._cursor_obj = None
self._limit = None
self._skip = None
-
+
def ensure_index(self, key_or_list):
"""Ensure that the given indexes are in place.
@@ -199,7 +201,7 @@ def _build_index_spec(cls, doc_cls, key_or_list):
return index_list
def __call__(self, q_obj=None, **query):
- """Filter the selected documents by calling the
+ """Filter the selected documents by calling the
:class:`~mongoengine.queryset.QuerySet` with a query.
:param q_obj: a :class:`~mongoengine.queryset.Q` object to be used in
@@ -213,7 +215,7 @@ def __call__(self, q_obj=None, **query):
query = QuerySet._transform_query(_doc_cls=self._document, **query)
self._query.update(query)
return self
-
+
def filter(self, *q_objs, **query):
"""An alias of :meth:`~mongoengine.queryset.QuerySet.__call__`
"""
@@ -253,11 +255,11 @@ def _cursor(self):
# Apply where clauses to cursor
if self._where_clause:
self._cursor_obj.where(self._where_clause)
-
+
# apply default ordering
if self._document._meta['ordering']:
self.order_by(*self._document._meta['ordering'])
-
+
return self._cursor_obj
@classmethod
@@ -337,8 +339,8 @@ def _transform_query(cls, _doc_cls=None, **query):
return mongo_query
def get(self, *q_objs, **query):
- """Retrieve the the matching object raising
- :class:`~mongoengine.queryset.MultipleObjectsReturned` or
+ """Retrieve the the matching object raising
+ :class:`~mongoengine.queryset.MultipleObjectsReturned` or
:class:`~mongoengine.queryset.DoesNotExist` exceptions if multiple or
no results are found.
"""
@@ -354,15 +356,15 @@ def get(self, *q_objs, **query):
def get_or_create(self, *q_objs, **query):
"""Retreive unique object or create, if it doesn't exist. Raises
- :class:`~mongoengine.queryset.MultipleObjectsReturned` if multiple
- results are found. A new document will be created if the document
+ :class:`~mongoengine.queryset.MultipleObjectsReturned` if multiple
+ results are found. A new document will be created if the document
doesn't exists; a dictionary of default values for the new document
may be provided as a keyword argument called :attr:`defaults`.
"""
defaults = query.get('defaults', {})
- if query.has_key('defaults'):
+ if 'defaults' in query:
del query['defaults']
-
+
self.__call__(*q_objs, **query)
count = self.count()
if count == 0:
@@ -439,6 +441,70 @@ def count(self):
def __len__(self):
return self.count()
+ def map_reduce(self, map_f, reduce_f, finalize_f=None, limit=None,
+ scope=None, keep_temp=False):
+ """Perform a map/reduce query using the current query spec
+ and ordering. While ``map_reduce`` respects ``QuerySet`` chaining,
+ it must be the last call made, as it does not return a maleable
+ ``QuerySet``.
+
+ See the :meth:`~mongoengine.tests.QuerySetTest.test_map_reduce`
+ and :meth:`~mongoengine.tests.QuerySetTest.test_map_advanced`
+ tests in ``tests.queryset.QuerySetTest`` for usage examples.
+
+ :param map_f: map function, as :class:`~pymongo.code.Code` or string
+ :param reduce_f: reduce function, as
+ :class:`~pymongo.code.Code` or string
+ :param finalize_f: finalize function, an optional function that
+ performs any post-reduction processing.
+ :param scope: values to insert into map/reduce global scope. Optional.
+ :param limit: number of objects from current query to provide
+ to map/reduce method
+ :param keep_temp: keep temporary table (boolean, default ``True``)
+
+ Returns an iterator yielding
+ :class:`~mongoengine.document.MapReduceDocument`.
+
+ .. note:: Map/Reduce requires server version **>= 1.1.1**. The PyMongo
+ :meth:`~pymongo.collection.Collection.map_reduce` helper requires
+ PyMongo version **>= 1.2**.
+
+ .. versionadded:: 0.3
+
+ """
+ from document import MapReduceDocument
+
+ if not hasattr(self._collection, "map_reduce"):
+ raise NotImplementedError("Requires MongoDB >= 1.1.1")
+
+ if not isinstance(map_f, pymongo.code.Code):
+ map_f = pymongo.code.Code(map_f)
+ if not isinstance(reduce_f, pymongo.code.Code):
+ reduce_f = pymongo.code.Code(reduce_f)
+
+ mr_args = {'query': self._query, 'keeptemp': keep_temp}
+
+ if finalize_f:
+ if not isinstance(finalize_f, pymongo.code.Code):
+ finalize_f = pymongo.code.Code(finalize_f)
+ mr_args['finalize'] = finalize_f
+
+ if scope:
+ mr_args['scope'] = scope
+
+ if limit:
+ mr_args['limit'] = limit
+
+ results = self._collection.map_reduce(map_f, reduce_f, **mr_args)
+ results = results.find()
+
+ if self._ordering:
+ results = results.sort(self._ordering)
+
+ for doc in results:
+ yield MapReduceDocument(self._document, self._collection,
+ doc['_id'], doc['value'])
+
def limit(self, n):
"""Limit the number of returned documents to `n`. This may also be
achieved using array-slicing syntax (e.g. ``User.objects[:5]``).
@@ -450,6 +516,7 @@ def limit(self, n):
else:
self._cursor.limit(n)
self._limit = n
+
# Return self to allow chaining
return self
@@ -523,13 +590,14 @@ def order_by(self, *keys):
direction = pymongo.DESCENDING
if key[0] in ('-', '+'):
key = key[1:]
- key_list.append((key, direction))
+ key_list.append((key, direction))
+ self._ordering = key_list
self._cursor.sort(key_list)
return self
-
+
def explain(self, format=False):
- """Return an explain plan record for the
+ """Return an explain plan record for the
:class:`~mongoengine.queryset.QuerySet`\ 's cursor.
:param format: format the plan before returning it
@@ -540,7 +608,7 @@ def explain(self, format=False):
import pprint
plan = pprint.pformat(plan)
return plan
-
+
def delete(self, safe=False):
"""Delete the documents matched by the query.
@@ -552,7 +620,7 @@ def delete(self, safe=False):
def _transform_update(cls, _doc_cls=None, **update):
"""Transform an update spec from Django-style format to Mongo format.
"""
- operators = ['set', 'unset', 'inc', 'dec', 'push', 'push_all', 'pull',
+ operators = ['set', 'unset', 'inc', 'dec', 'push', 'push_all', 'pull',
'pull_all']
mongo_update = {}
@@ -661,8 +729,8 @@ def exec_js(self, code, *fields, **options):
"""Execute a Javascript function on the server. A list of fields may be
provided, which will be translated to their correct names and supplied
as the arguments to the function. A few extra variables are added to
- the function's scope: ``collection``, which is the name of the
- collection in use; ``query``, which is an object representing the
+ the function's scope: ``collection``, which is the name of the
+ collection in use; ``query``, which is an object representing the
current query; and ``options``, which is an object containing any
options specified as keyword arguments.
@@ -676,7 +744,7 @@ def exec_js(self, code, *fields, **options):
:param code: a string of Javascript code to execute
:param fields: fields that you will be using in your function, which
will be passed in to your function as arguments
- :param options: options that you want available to the function
+ :param options: options that you want available to the function
(accessed in Javascript through the ``options`` object)
"""
code = self._sub_js_fields(code)
@@ -693,7 +761,7 @@ def exec_js(self, code, *fields, **options):
query = self._query
if self._where_clause:
query['$where'] = self._where_clause
-
+
scope['query'] = query
code = pymongo.code.Code(code, scope=scope)
@@ -741,7 +809,7 @@ def average(self, field):
def item_frequencies(self, list_field, normalize=False):
"""Returns a dictionary of all items present in a list field across
the whole queried set of documents, and their corresponding frequency.
- This is useful for generating tag clouds, or searching documents.
+ This is useful for generating tag clouds, or searching documents.
:param list_field: the list field to use
:param normalize: normalize the results so they add to 1.0
@@ -791,7 +859,7 @@ def __init__(self, manager_func=None):
self._collection = None
def __get__(self, instance, owner):
- """Descriptor for instantiating a new QuerySet object when
+ """Descriptor for instantiating a new QuerySet object when
Document.objects is accessed.
"""
if instance is not None:
@@ -810,7 +878,7 @@ def __get__(self, instance, owner):
if collection in db.collection_names():
self._collection = db[collection]
- # The collection already exists, check if its capped
+ # The collection already exists, check if its capped
# options match the specified capped options
options = self._collection.options()
if options.get('max') != max_documents or \
@@ -826,7 +894,7 @@ def __get__(self, instance, owner):
self._collection = db.create_collection(collection, opts)
else:
self._collection = db[collection]
-
+
# owner is the document that contains the QuerySetManager
queryset = QuerySet(owner, self._collection)
if self._manager_func:
@@ -836,6 +904,7 @@ def __get__(self, instance, owner):
queryset = self._manager_func(owner, queryset)
return queryset
+
def queryset_manager(func):
"""Decorator that allows you to define custom QuerySet managers on
:class:`~mongoengine.Document` classes. The manager must be a function that
252 tests/queryset.py
View
@@ -1,14 +1,17 @@
+# -*- coding: utf-8 -*-
+
+
import unittest
import pymongo
-from datetime import datetime
+from datetime import datetime, timedelta
-from mongoengine.queryset import (QuerySet, MultipleObjectsReturned,
+from mongoengine.queryset import (QuerySet, MultipleObjectsReturned,
DoesNotExist)
from mongoengine import *
class QuerySetTest(unittest.TestCase):
-
+
def setUp(self):
connect(db='mongoenginetest')
@@ -16,12 +19,12 @@ class Person(Document):
name = StringField()
age = IntField()
self.Person = Person
-
+
def test_initialisation(self):
"""Ensure that a QuerySet is correctly initialised by QuerySetManager.
"""
self.assertTrue(isinstance(self.Person.objects, QuerySet))
- self.assertEqual(self.Person.objects._collection.name,
+ self.assertEqual(self.Person.objects._collection.name,
self.Person._meta['collection'])
self.assertTrue(isinstance(self.Person.objects._collection,
pymongo.collection.Collection))
@@ -31,15 +34,15 @@ def test_transform_query(self):
"""
self.assertEqual(QuerySet._transform_query(name='test', age=30),
{'name': 'test', 'age': 30})
- self.assertEqual(QuerySet._transform_query(age__lt=30),
+ self.assertEqual(QuerySet._transform_query(age__lt=30),
{'age': {'$lt': 30}})
self.assertEqual(QuerySet._transform_query(age__gt=20, age__lt=50),
{'age': {'$gt': 20, '$lt': 50}})
self.assertEqual(QuerySet._transform_query(age=20, age__gt=50),
{'age': 20})
- self.assertEqual(QuerySet._transform_query(friend__age__gte=30),
+ self.assertEqual(QuerySet._transform_query(friend__age__gte=30),
{'friend.age': {'$gte': 30}})
- self.assertEqual(QuerySet._transform_query(name__exists=True),
+ self.assertEqual(QuerySet._transform_query(name__exists=True),
{'name': {'$exists': True}})
def test_find(self):
@@ -134,7 +137,7 @@ def test_find_one(self):
self.assertEqual(person.name, "User B")
self.assertRaises(IndexError, self.Person.objects.__getitem__, 2)
-
+
# Find a document using just the object id
person = self.Person.objects.with_id(person1.id)
self.assertEqual(person.name, "User A")
@@ -170,7 +173,7 @@ def test_get_or_create(self):
person2.save()
# Retrieve the first person from the database
- self.assertRaises(MultipleObjectsReturned,
+ self.assertRaises(MultipleObjectsReturned,
self.Person.objects.get_or_create)
# Use a query to filter the people found to just person2
@@ -256,36 +259,36 @@ def test_filter_chaining(self):
"""Ensure filters can be chained together.
"""
from datetime import datetime
-
+
class BlogPost(Document):
title = StringField()
is_published = BooleanField()
published_date = DateTimeField()
-
+
@queryset_manager
def published(doc_cls, queryset):
return queryset(is_published=True)
-
- blog_post_1 = BlogPost(title="Blog Post #1",
+
+ blog_post_1 = BlogPost(title="Blog Post #1",
is_published = True,
published_date=datetime(2010, 1, 5, 0, 0 ,0))
- blog_post_2 = BlogPost(title="Blog Post #2",
+ blog_post_2 = BlogPost(title="Blog Post #2",
is_published = True,
published_date=datetime(2010, 1, 6, 0, 0 ,0))
- blog_post_3 = BlogPost(title="Blog Post #3",
+ blog_post_3 = BlogPost(title="Blog Post #3",
is_published = True,
published_date=datetime(2010, 1, 7, 0, 0 ,0))
blog_post_1.save()
blog_post_2.save()
blog_post_3.save()
-
+
# find all published blog posts before 2010-01-07
published_posts = BlogPost.published()
published_posts = published_posts.filter(
published_date__lt=datetime(2010, 1, 7, 0, 0 ,0))
self.assertEqual(published_posts.count(), 2)
-
+
BlogPost.drop_collection()
def test_ordering(self):
@@ -301,22 +304,22 @@ class BlogPost(Document):
BlogPost.drop_collection()
- blog_post_1 = BlogPost(title="Blog Post #1",
+ blog_post_1 = BlogPost(title="Blog Post #1",
published_date=datetime(2010, 1, 5, 0, 0 ,0))
- blog_post_2 = BlogPost(title="Blog Post #2",
+ blog_post_2 = BlogPost(title="Blog Post #2",
published_date=datetime(2010, 1, 6, 0, 0 ,0))
- blog_post_3 = BlogPost(title="Blog Post #3",
+ blog_post_3 = BlogPost(title="Blog Post #3",
published_date=datetime(2010, 1, 7, 0, 0 ,0))
blog_post_1.save()
blog_post_2.save()
blog_post_3.save()
-
+
# get the "first" BlogPost using default ordering
# from BlogPost.meta.ordering
- latest_post = BlogPost.objects.first()
+ latest_post = BlogPost.objects.first()
self.assertEqual(latest_post.title, "Blog Post #3")
-
+
# override default ordering, order BlogPosts by "published_date"
first_post = BlogPost.objects.order_by("+published_date").first()
self.assertEqual(first_post.title, "Blog Post #1")
@@ -375,7 +378,7 @@ class BlogPost(Document):
result = BlogPost.objects.first()
self.assertTrue(isinstance(result.author, User))
self.assertEqual(result.author.name, 'Test User')
-
+
BlogPost.drop_collection()
def test_find_dict_item(self):
@@ -442,7 +445,7 @@ class BlogPost(Document):
self.Person(name='user2', age=20).save()
self.Person(name='user3', age=30).save()
self.Person(name='user4', age=40).save()
-
+
self.assertEqual(len(self.Person.objects(Q(age__in=[20]))), 2)
self.assertEqual(len(self.Person.objects(Q(age__in=[20, 30]))), 3)
@@ -545,17 +548,17 @@ class BlogPost(Document):
return comments;
}
"""
-
+
sub_code = BlogPost.objects._sub_js_fields(code)
- code_chunks = ['doc["cmnts"];', 'doc["doc-name"],',
+ code_chunks = ['doc["cmnts"];', 'doc["doc-name"],',
'doc["cmnts"][i]["body"]']
for chunk in code_chunks:
self.assertTrue(chunk in sub_code)
results = BlogPost.objects.exec_js(code)
expected_results = [
- {u'comment': u'cool', u'document': u'post1'},
- {u'comment': u'yay', u'document': u'post1'},
+ {u'comment': u'cool', u'document': u'post1'},
+ {u'comment': u'yay', u'document': u'post1'},
{u'comment': u'nice stuff', u'document': u'post2'},
]
self.assertEqual(results, expected_results)
@@ -627,10 +630,167 @@ def test_order_by(self):
names = [p.name for p in self.Person.objects.order_by('age')]
self.assertEqual(names, ['User A', 'User C', 'User B'])
-
+
ages = [p.age for p in self.Person.objects.order_by('-name')]
self.assertEqual(ages, [30, 40, 20])
+ def test_map_reduce(self):
+ """Ensure map/reduce is both mapping and reducing.
+ """
+ class BlogPost(Document):
+ title = StringField()
+ tags = ListField(StringField())
+
+ BlogPost.drop_collection()
+
+ BlogPost(title="Post #1", tags=['music', 'film', 'print']).save()
+ BlogPost(title="Post #2", tags=['music', 'film']).save()
+ BlogPost(title="Post #3", tags=['film', 'photography']).save()
+
+ map_f = """
+ function() {
+ this.tags.forEach(function(tag) {
+ emit(tag, 1);
+ });
+ }
+ """
+
+ reduce_f = """
+ function(key, values) {
+ var total = 0;
+ for(var i=0; i<values.length; i++) {
+ total += values[i];
+ }
+ return total;
+ }
+ """
+
+ # run a map/reduce operation spanning all posts
+ results = BlogPost.objects.map_reduce(map_f, reduce_f)
+ results = list(results)
+ self.assertEqual(len(results), 4)
+
+ music = filter(lambda r: r.key == "music", results)[0]
+ self.assertEqual(music.value, 2)
+
+ film = filter(lambda r: r.key == "film", results)[0]
+ self.assertEqual(film.value, 3)
+
+ BlogPost.drop_collection()
+
+ def test_map_reduce_finalize(self):
+ """Ensure that map, reduce, and finalize run and introduce "scope"
+ by simulating "hotness" ranking with Reddit algorithm.
+ """
+ from time import mktime
+
+ class Link(Document):
+ title = StringField()
+ up_votes = IntField()
+ down_votes = IntField()
+ submitted = DateTimeField()
+
+ Link.drop_collection()
+
+ now = datetime.utcnow()
+
+ # Note: Test data taken from a custom Reddit homepage on
+ # Fri, 12 Feb 2010 14:36:00 -0600. Link ordering should
+ # reflect order of insertion below, but is not influenced
+ # by insertion order.
+ Link(title = "Google Buzz auto-followed a woman's abusive ex ...",
+ up_votes = 1079,
+ down_votes = 553,
+ submitted = now-timedelta(hours=4)).save()
+ Link(title = "We did it! Barbie is a computer engineer.",
+ up_votes = 481,
+ down_votes = 124,
+ submitted = now-timedelta(hours=2)).save()
+ Link(title = "This Is A Mosquito Getting Killed By A Laser",
+ up_votes = 1446,
+ down_votes = 530,
+ submitted=now-timedelta(hours=13)).save()
+ Link(title = "Arabic flashcards land physics student in jail.",
+ up_votes = 215,
+ down_votes = 105,
+ submitted = now-timedelta(hours=6)).save()
+ Link(title = "The Burger Lab: Presenting, the Flood Burger",
+ up_votes = 48,
+ down_votes = 17,
+ submitted = now-timedelta(hours=5)).save()
+ Link(title="How to see polarization with the naked eye",
+ up_votes = 74,
+ down_votes = 13,
+ submitted = now-timedelta(hours=10)).save()
+
+ map_f = """
+ function() {
+ emit(this._id, {up_delta: this.up_votes - this.down_votes,
+ sub_date: this.submitted.getTime() / 1000})
+ }
+ """
+
+ reduce_f = """
+ function(key, values) {
+ data = values[0];
+
+ x = data.up_delta;
+
+ // calculate time diff between reddit epoch and submission
+ sec_since_epoch = data.sub_date - reddit_epoch;
+
+ // calculate 'Y'
+ if(x > 0) {
+ y = 1;
+ } else if (x = 0) {
+ y = 0;
+ } else {
+ y = -1;
+ }
+
+ // calculate 'Z', the maximal value
+ if(Math.abs(x) >= 1) {
+ z = Math.abs(x);
+ } else {
+ z = 1;
+ }
+
+ return {x: x, y: y, z: z, t_s: sec_since_epoch};
+ }
+ """
+
+ finalize_f = """
+ function(key, value) {
+ // f(sec_since_epoch,y,z) = log10(z) + ((y*sec_since_epoch) / 45000)
+ z_10 = Math.log(value.z) / Math.log(10);
+ weight = z_10 + ((value.y * value.t_s) / 45000);
+ return weight;
+ }
+ """
+
+ # provide the reddit epoch (used for ranking) as a variable available
+ # to all phases of the map/reduce operation: map, reduce, and finalize.
+ reddit_epoch = mktime(datetime(2005, 12, 8, 7, 46, 43).timetuple())
+ scope = {'reddit_epoch': reddit_epoch}
+
+ # run a map/reduce operation across all links. ordering is set
+ # to "-value", which orders the "weight" value returned from
+ # "finalize_f" in descending order.
+ results = Link.objects.order_by("-value")
+ results = results.map_reduce(map_f,
+ reduce_f,
+ finalize_f=finalize_f,
+ scope=scope)
+ results = list(results)
+
+ # assert troublesome Buzz article is ranked 1st
+ self.assertTrue(results[0].object.title.startswith("Google Buzz"))
+
+ # assert laser vision is ranked last
+ self.assertTrue(results[-1].object.title.startswith("How to see"))
+
+ Link.drop_collection()
+
def test_item_frequencies(self):
"""Ensure that item frequencies are properly generated from lists.
"""
@@ -727,20 +887,20 @@ class BlogPost(Document):
title = StringField(name='postTitle')
comments = ListField(EmbeddedDocumentField(Comment),
name='postComments')
-
+
BlogPost.drop_collection()
data = {'title': 'Post 1', 'comments': [Comment(content='test')]}
BlogPost(**data).save()
- self.assertTrue('postTitle' in
+ self.assertTrue('postTitle' in
BlogPost.objects(title=data['title'])._query)
- self.assertFalse('title' in
+ self.assertFalse('title' in
BlogPost.objects(title=data['title'])._query)
self.assertEqual(len(BlogPost.objects(title=data['title'])), 1)
- self.assertTrue('postComments.commentContent' in
+ self.assertTrue('postComments.commentContent' in
BlogPost.objects(comments__content='test')._query)
self.assertEqual(len(BlogPost.objects(comments__content='test')), 1)
@@ -761,7 +921,7 @@ class BlogPost(Document):
post.save()
# Test that query may be performed by providing a document as a value
- # while using a ReferenceField's name - the document should be
+ # while using a ReferenceField's name - the document should be
# converted to an DBRef, which is legal, unlike a Document object
post_obj = BlogPost.objects(author=person).first()
self.assertEqual(post.id, post_obj.id)
@@ -823,13 +983,13 @@ class BlogPost(Document):
self.assertFalse([('_types', 1)] in info.values())
BlogPost.drop_collection()
-
+
def test_bulk(self):
"""Ensure bulk querying by object id returns a proper dict.
"""
class BlogPost(Document):
title = StringField()
-
+
BlogPost.drop_collection()
post_1 = BlogPost(title="Post #1")
@@ -843,20 +1003,20 @@ class BlogPost(Document):
post_3.save()
post_4.save()
post_5.save()
-
+
ids = [post_1.id, post_2.id, post_5.id]
objects = BlogPost.objects.in_bulk(ids)
-
+
self.assertEqual(len(objects), 3)
self.assertTrue(post_1.id in objects)
self.assertTrue(post_2.id in objects)
self.assertTrue(post_5.id in objects)
-
+
self.assertTrue(objects[post_1.id].title == post_1.title)
self.assertTrue(objects[post_2.id].title == post_2.title)
- self.assertTrue(objects[post_5.id].title == post_5.title)
-
+ self.assertTrue(objects[post_5.id].title == post_5.title)
+
BlogPost.drop_collection()
def tearDown(self):
@@ -864,7 +1024,7 @@ def tearDown(self):
class QTest(unittest.TestCase):
-
+
def test_or_and(self):
"""Ensure that Q objects may be combined correctly.
"""
@@ -888,8 +1048,8 @@ def test_item_query_as_js(self):
examples = [
({'name': 'test'}, 'this.name == i0f0', {'i0f0': 'test'}),
({'age': {'$gt': 18}}, 'this.age > i0f0o0', {'i0f0o0': 18}),
- ({'name': 'test', 'age': {'$gt': 18, '$lte': 65}},
- 'this.age <= i0f0o0 && this.age > i0f0o1 && this.name == i0f1',
+ ({'name': 'test', 'age': {'$gt': 18, '$lte': 65}},
+ 'this.age <= i0f0o0 && this.age > i0f0o1 && this.name == i0f1',
{'i0f0o0': 65, 'i0f0o1': 18, 'i0f1': 'test'}),
]
for item, js, scope in examples:
Please sign in to comment.
Something went wrong with that request. Please try again.