379 lines
19 KiB
Python
379 lines
19 KiB
Python
# coding: utf-8
|
|
import pickle
|
|
|
|
from django.db import connection, models, DEFAULT_DB_ALIAS
|
|
from django.conf import settings
|
|
|
|
class Author(models.Model):
|
|
name = models.CharField(max_length=100)
|
|
age = models.IntegerField()
|
|
friends = models.ManyToManyField('self', blank=True)
|
|
|
|
def __unicode__(self):
|
|
return self.name
|
|
|
|
class Publisher(models.Model):
|
|
name = models.CharField(max_length=255)
|
|
num_awards = models.IntegerField()
|
|
|
|
def __unicode__(self):
|
|
return self.name
|
|
|
|
class Book(models.Model):
|
|
isbn = models.CharField(max_length=9)
|
|
name = models.CharField(max_length=255)
|
|
pages = models.IntegerField()
|
|
rating = models.FloatField()
|
|
price = models.DecimalField(decimal_places=2, max_digits=6)
|
|
authors = models.ManyToManyField(Author)
|
|
contact = models.ForeignKey(Author, related_name='book_contact_set')
|
|
publisher = models.ForeignKey(Publisher)
|
|
pubdate = models.DateField()
|
|
|
|
class Meta:
|
|
ordering = ('name',)
|
|
|
|
def __unicode__(self):
|
|
return self.name
|
|
|
|
class Store(models.Model):
|
|
name = models.CharField(max_length=255)
|
|
books = models.ManyToManyField(Book)
|
|
original_opening = models.DateTimeField()
|
|
friday_night_closing = models.TimeField()
|
|
|
|
def __unicode__(self):
|
|
return self.name
|
|
|
|
class Entries(models.Model):
|
|
EntryID = models.AutoField(primary_key=True, db_column='Entry ID')
|
|
Entry = models.CharField(unique=True, max_length=50)
|
|
Exclude = models.BooleanField()
|
|
|
|
class Clues(models.Model):
|
|
ID = models.AutoField(primary_key=True)
|
|
EntryID = models.ForeignKey(Entries, verbose_name='Entry', db_column = 'Entry ID')
|
|
Clue = models.CharField(max_length=150)
|
|
|
|
class HardbackBook(Book):
|
|
weight = models.FloatField()
|
|
|
|
def __unicode__(self):
|
|
return "%s (hardback): %s" % (self.name, self.weight)
|
|
|
|
__test__ = {'API_TESTS': """
|
|
>>> from django.core import management
|
|
>>> from django.db.models import get_app, F
|
|
|
|
# Reset the database representation of this app.
|
|
# This will return the database to a clean initial state.
|
|
>>> management.call_command('flush', verbosity=0, interactive=False)
|
|
|
|
>>> from django.db.models import Avg, Sum, Count, Max, Min, StdDev, Variance
|
|
|
|
# Ordering requests are ignored
|
|
>>> Author.objects.all().order_by('name').aggregate(Avg('age'))
|
|
{'age__avg': 37.4...}
|
|
|
|
# Implicit ordering is also ignored
|
|
>>> Book.objects.all().aggregate(Sum('pages'))
|
|
{'pages__sum': 3703}
|
|
|
|
# Baseline results
|
|
>>> Book.objects.all().aggregate(Sum('pages'), Avg('pages'))
|
|
{'pages__sum': 3703, 'pages__avg': 617.1...}
|
|
|
|
# Empty values query doesn't affect grouping or results
|
|
>>> Book.objects.all().values().aggregate(Sum('pages'), Avg('pages'))
|
|
{'pages__sum': 3703, 'pages__avg': 617.1...}
|
|
|
|
# Aggregate overrides extra selected column
|
|
>>> Book.objects.all().extra(select={'price_per_page' : 'price / pages'}).aggregate(Sum('pages'))
|
|
{'pages__sum': 3703}
|
|
|
|
# Annotations get combined with extra select clauses
|
|
>>> sorted((k,v) for k,v in Book.objects.all().annotate(mean_auth_age=Avg('authors__age')).extra(select={'manufacture_cost' : 'price * .5'}).get(pk=2).__dict__.items() if k != '_state')
|
|
[('contact_id', 3), ('id', 2), ('isbn', u'067232959'), ('manufacture_cost', ...11.545...), ('mean_auth_age', 45.0), ('name', u'Sams Teach Yourself Django in 24 Hours'), ('pages', 528), ('price', Decimal("23.09")), ('pubdate', datetime.date(2008, 3, 3)), ('publisher_id', 2), ('rating', 3.0)]
|
|
|
|
# Order of the annotate/extra in the query doesn't matter
|
|
>>> sorted((k,v) for k,v in Book.objects.all().extra(select={'manufacture_cost' : 'price * .5'}).annotate(mean_auth_age=Avg('authors__age')).get(pk=2).__dict__.items()if k != '_state')
|
|
[('contact_id', 3), ('id', 2), ('isbn', u'067232959'), ('manufacture_cost', ...11.545...), ('mean_auth_age', 45.0), ('name', u'Sams Teach Yourself Django in 24 Hours'), ('pages', 528), ('price', Decimal("23.09")), ('pubdate', datetime.date(2008, 3, 3)), ('publisher_id', 2), ('rating', 3.0)]
|
|
|
|
# Values queries can be combined with annotate and extra
|
|
>>> sorted((k,v) for k,v in Book.objects.all().annotate(mean_auth_age=Avg('authors__age')).extra(select={'manufacture_cost' : 'price * .5'}).values().get(pk=2).items()if k != '_state')
|
|
[('contact_id', 3), ('id', 2), ('isbn', u'067232959'), ('manufacture_cost', ...11.545...), ('mean_auth_age', 45.0), ('name', u'Sams Teach Yourself Django in 24 Hours'), ('pages', 528), ('price', Decimal("23.09")), ('pubdate', datetime.date(2008, 3, 3)), ('publisher_id', 2), ('rating', 3.0)]
|
|
|
|
# The order of the (empty) values, annotate and extra clauses doesn't matter
|
|
>>> sorted((k,v) for k,v in Book.objects.all().values().annotate(mean_auth_age=Avg('authors__age')).extra(select={'manufacture_cost' : 'price * .5'}).get(pk=2).items()if k != '_state')
|
|
[('contact_id', 3), ('id', 2), ('isbn', u'067232959'), ('manufacture_cost', ...11.545...), ('mean_auth_age', 45.0), ('name', u'Sams Teach Yourself Django in 24 Hours'), ('pages', 528), ('price', Decimal("23.09")), ('pubdate', datetime.date(2008, 3, 3)), ('publisher_id', 2), ('rating', 3.0)]
|
|
|
|
# If the annotation precedes the values clause, it won't be included
|
|
# unless it is explicitly named
|
|
>>> sorted(Book.objects.all().annotate(mean_auth_age=Avg('authors__age')).extra(select={'price_per_page' : 'price / pages'}).values('name').get(pk=1).items())
|
|
[('name', u'The Definitive Guide to Django: Web Development Done Right')]
|
|
|
|
>>> sorted(Book.objects.all().annotate(mean_auth_age=Avg('authors__age')).extra(select={'price_per_page' : 'price / pages'}).values('name','mean_auth_age').get(pk=1).items())
|
|
[('mean_auth_age', 34.5), ('name', u'The Definitive Guide to Django: Web Development Done Right')]
|
|
|
|
# If an annotation isn't included in the values, it can still be used in a filter
|
|
>>> Book.objects.annotate(n_authors=Count('authors')).values('name').filter(n_authors__gt=2)
|
|
[{'name': u'Python Web Development with Django'}]
|
|
|
|
# The annotations are added to values output if values() precedes annotate()
|
|
>>> sorted(Book.objects.all().values('name').annotate(mean_auth_age=Avg('authors__age')).extra(select={'price_per_page' : 'price / pages'}).get(pk=1).items())
|
|
[('mean_auth_age', 34.5), ('name', u'The Definitive Guide to Django: Web Development Done Right')]
|
|
|
|
# Check that all of the objects are getting counted (allow_nulls) and that values respects the amount of objects
|
|
>>> len(Author.objects.all().annotate(Avg('friends__age')).values())
|
|
9
|
|
|
|
# Check that consecutive calls to annotate accumulate in the query
|
|
>>> Book.objects.values('price').annotate(oldest=Max('authors__age')).order_by('oldest', 'price').annotate(Max('publisher__num_awards'))
|
|
[{'price': Decimal("30..."), 'oldest': 35, 'publisher__num_awards__max': 3}, {'price': Decimal("29.69"), 'oldest': 37, 'publisher__num_awards__max': 7}, {'price': Decimal("23.09"), 'oldest': 45, 'publisher__num_awards__max': 1}, {'price': Decimal("75..."), 'oldest': 57, 'publisher__num_awards__max': 9}, {'price': Decimal("82.8..."), 'oldest': 57, 'publisher__num_awards__max': 7}]
|
|
|
|
# Aggregates can be composed over annotations.
|
|
# The return type is derived from the composed aggregate
|
|
>>> Book.objects.all().annotate(num_authors=Count('authors__id')).aggregate(Max('pages'), Max('price'), Sum('num_authors'), Avg('num_authors'))
|
|
{'num_authors__sum': 10, 'num_authors__avg': 1.66..., 'pages__max': 1132, 'price__max': Decimal("82.80")}
|
|
|
|
# Bad field requests in aggregates are caught and reported
|
|
>>> Book.objects.all().aggregate(num_authors=Count('foo'))
|
|
Traceback (most recent call last):
|
|
...
|
|
FieldError: Cannot resolve keyword 'foo' into field. Choices are: authors, contact, hardbackbook, id, isbn, name, pages, price, pubdate, publisher, rating, store
|
|
|
|
>>> Book.objects.all().annotate(num_authors=Count('foo'))
|
|
Traceback (most recent call last):
|
|
...
|
|
FieldError: Cannot resolve keyword 'foo' into field. Choices are: authors, contact, hardbackbook, id, isbn, name, pages, price, pubdate, publisher, rating, store
|
|
|
|
>>> Book.objects.all().annotate(num_authors=Count('authors__id')).aggregate(Max('foo'))
|
|
Traceback (most recent call last):
|
|
...
|
|
FieldError: Cannot resolve keyword 'foo' into field. Choices are: authors, contact, hardbackbook, id, isbn, name, pages, price, pubdate, publisher, rating, store, num_authors
|
|
|
|
# Old-style count aggregations can be mixed with new-style
|
|
>>> Book.objects.annotate(num_authors=Count('authors')).count()
|
|
6
|
|
|
|
# Non-ordinal, non-computed Aggregates over annotations correctly inherit
|
|
# the annotation's internal type if the annotation is ordinal or computed
|
|
>>> Book.objects.annotate(num_authors=Count('authors')).aggregate(Max('num_authors'))
|
|
{'num_authors__max': 3}
|
|
|
|
>>> Publisher.objects.annotate(avg_price=Avg('book__price')).aggregate(Max('avg_price'))
|
|
{'avg_price__max': 75.0...}
|
|
|
|
# Aliases are quoted to protected aliases that might be reserved names
|
|
>>> Book.objects.aggregate(number=Max('pages'), select=Max('pages'))
|
|
{'number': 1132, 'select': 1132}
|
|
|
|
# Regression for #10064: select_related() plays nice with aggregates
|
|
>>> sorted(Book.objects.select_related('publisher').annotate(num_authors=Count('authors')).values()[0].iteritems())
|
|
[('contact_id', 8), ('id', 5), ('isbn', u'013790395'), ('name', u'Artificial Intelligence: A Modern Approach'), ('num_authors', 2), ('pages', 1132), ('price', Decimal("82.8...")), ('pubdate', datetime.date(1995, 1, 15)), ('publisher_id', 3), ('rating', 4.0)]
|
|
|
|
# Regression for #10010: exclude on an aggregate field is correctly negated
|
|
>>> len(Book.objects.annotate(num_authors=Count('authors')))
|
|
6
|
|
>>> len(Book.objects.annotate(num_authors=Count('authors')).filter(num_authors__gt=2))
|
|
1
|
|
>>> len(Book.objects.annotate(num_authors=Count('authors')).exclude(num_authors__gt=2))
|
|
5
|
|
|
|
>>> len(Book.objects.annotate(num_authors=Count('authors')).filter(num_authors__lt=3).exclude(num_authors__lt=2))
|
|
2
|
|
>>> len(Book.objects.annotate(num_authors=Count('authors')).exclude(num_authors__lt=2).filter(num_authors__lt=3))
|
|
2
|
|
|
|
# Aggregates can be used with F() expressions
|
|
# ... where the F() is pushed into the HAVING clause
|
|
>>> Publisher.objects.annotate(num_books=Count('book')).filter(num_books__lt=F('num_awards')/2).order_by('name').values('name','num_books','num_awards')
|
|
[{'num_books': 1, 'name': u'Morgan Kaufmann', 'num_awards': 9}, {'num_books': 2, 'name': u'Prentice Hall', 'num_awards': 7}]
|
|
|
|
>>> Publisher.objects.annotate(num_books=Count('book')).exclude(num_books__lt=F('num_awards')/2).order_by('name').values('name','num_books','num_awards')
|
|
[{'num_books': 2, 'name': u'Apress', 'num_awards': 3}, {'num_books': 0, 'name': u"Jonno's House of Books", 'num_awards': 0}, {'num_books': 1, 'name': u'Sams', 'num_awards': 1}]
|
|
|
|
# ... and where the F() references an aggregate
|
|
>>> Publisher.objects.annotate(num_books=Count('book')).filter(num_awards__gt=2*F('num_books')).order_by('name').values('name','num_books','num_awards')
|
|
[{'num_books': 1, 'name': u'Morgan Kaufmann', 'num_awards': 9}, {'num_books': 2, 'name': u'Prentice Hall', 'num_awards': 7}]
|
|
|
|
>>> Publisher.objects.annotate(num_books=Count('book')).exclude(num_books__lt=F('num_awards')/2).order_by('name').values('name','num_books','num_awards')
|
|
[{'num_books': 2, 'name': u'Apress', 'num_awards': 3}, {'num_books': 0, 'name': u"Jonno's House of Books", 'num_awards': 0}, {'num_books': 1, 'name': u'Sams', 'num_awards': 1}]
|
|
|
|
# Tests on fields with non-default table and column names.
|
|
>>> Clues.objects.values('EntryID__Entry').annotate(Appearances=Count('EntryID'), Distinct_Clues=Count('Clue', distinct=True))
|
|
[]
|
|
|
|
>>> Entries.objects.annotate(clue_count=Count('clues__ID'))
|
|
[]
|
|
|
|
# Regression for #10089: Check handling of empty result sets with aggregates
|
|
>>> Book.objects.filter(id__in=[]).count()
|
|
0
|
|
|
|
>>> Book.objects.filter(id__in=[]).aggregate(num_authors=Count('authors'), avg_authors=Avg('authors'), max_authors=Max('authors'), max_price=Max('price'), max_rating=Max('rating'))
|
|
{'max_authors': None, 'max_rating': None, 'num_authors': 0, 'avg_authors': None, 'max_price': None}
|
|
|
|
>>> list(Publisher.objects.filter(pk=5).annotate(num_authors=Count('book__authors'), avg_authors=Avg('book__authors'), max_authors=Max('book__authors'), max_price=Max('book__price'), max_rating=Max('book__rating')).values()) == [{'max_authors': None, 'name': u"Jonno's House of Books", 'num_awards': 0, 'max_price': None, 'num_authors': 0, 'max_rating': None, 'id': 5, 'avg_authors': None}]
|
|
True
|
|
|
|
# Regression for #10113 - Fields mentioned in order_by() must be included in the GROUP BY.
|
|
# This only becomes a problem when the order_by introduces a new join.
|
|
>>> Book.objects.annotate(num_authors=Count('authors')).order_by('publisher__name', 'name')
|
|
[<Book: Practical Django Projects>, <Book: The Definitive Guide to Django: Web Development Done Right>, <Book: Paradigms of Artificial Intelligence Programming: Case Studies in Common Lisp>, <Book: Artificial Intelligence: A Modern Approach>, <Book: Python Web Development with Django>, <Book: Sams Teach Yourself Django in 24 Hours>]
|
|
|
|
# Regression for #10127 - Empty select_related() works with annotate
|
|
>>> books = Book.objects.all().filter(rating__lt=4.5).select_related().annotate(Avg('authors__age'))
|
|
>>> sorted([(b.name, b.authors__age__avg, b.publisher.name, b.contact.name) for b in books])
|
|
[(u'Artificial Intelligence: A Modern Approach', 51.5, u'Prentice Hall', u'Peter Norvig'), (u'Practical Django Projects', 29.0, u'Apress', u'James Bennett'), (u'Python Web Development with Django', 30.3..., u'Prentice Hall', u'Jeffrey Forcier'), (u'Sams Teach Yourself Django in 24 Hours', 45.0, u'Sams', u'Brad Dayley')]
|
|
|
|
# Regression for #10132 - If the values() clause only mentioned extra(select=) columns, those columns are used for grouping
|
|
>>> Book.objects.extra(select={'pub':'publisher_id'}).values('pub').annotate(Count('id')).order_by('pub')
|
|
[{'pub': 1, 'id__count': 2}, {'pub': 2, 'id__count': 1}, {'pub': 3, 'id__count': 2}, {'pub': 4, 'id__count': 1}]
|
|
|
|
>>> Book.objects.extra(select={'pub':'publisher_id','foo':'pages'}).values('pub').annotate(Count('id')).order_by('pub')
|
|
[{'pub': 1, 'id__count': 2}, {'pub': 2, 'id__count': 1}, {'pub': 3, 'id__count': 2}, {'pub': 4, 'id__count': 1}]
|
|
|
|
# Regression for #10182 - Queries with aggregate calls are correctly realiased when used in a subquery
|
|
>>> ids = Book.objects.filter(pages__gt=100).annotate(n_authors=Count('authors')).filter(n_authors__gt=2).order_by('n_authors')
|
|
>>> Book.objects.filter(id__in=ids)
|
|
[<Book: Python Web Development with Django>]
|
|
|
|
# Regression for #10197 -- Queries with aggregates can be pickled.
|
|
# First check that pickling is possible at all. No crash = success
|
|
>>> qs = Book.objects.annotate(num_authors=Count('authors'))
|
|
>>> out = pickle.dumps(qs)
|
|
|
|
# Then check that the round trip works.
|
|
>>> query = qs.query.get_compiler(qs.db).as_sql()[0]
|
|
>>> select_fields = qs.query.select_fields
|
|
>>> query2 = pickle.loads(pickle.dumps(qs))
|
|
>>> query2.query.get_compiler(query2.db).as_sql()[0] == query
|
|
True
|
|
>>> query2.query.select_fields = select_fields
|
|
|
|
# Regression for #10199 - Aggregate calls clone the original query so the original query can still be used
|
|
>>> books = Book.objects.all()
|
|
>>> _ = books.aggregate(Avg('authors__age'))
|
|
>>> books.all()
|
|
[<Book: Artificial Intelligence: A Modern Approach>, <Book: Paradigms of Artificial Intelligence Programming: Case Studies in Common Lisp>, <Book: Practical Django Projects>, <Book: Python Web Development with Django>, <Book: Sams Teach Yourself Django in 24 Hours>, <Book: The Definitive Guide to Django: Web Development Done Right>]
|
|
|
|
# Regression for #10248 - Annotations work with DateQuerySets
|
|
>>> Book.objects.annotate(num_authors=Count('authors')).filter(num_authors=2).dates('pubdate', 'day')
|
|
[datetime.datetime(1995, 1, 15, 0, 0), datetime.datetime(2007, 12, 6, 0, 0)]
|
|
|
|
# Regression for #10290 - extra selects with parameters can be used for
|
|
# grouping.
|
|
>>> qs = Book.objects.all().annotate(mean_auth_age=Avg('authors__age')).extra(select={'sheets' : '(pages + %s) / %s'}, select_params=[1, 2]).order_by('sheets').values('sheets')
|
|
>>> [int(x['sheets']) for x in qs]
|
|
[150, 175, 224, 264, 473, 566]
|
|
|
|
# Regression for 10425 - annotations don't get in the way of a count() clause
|
|
>>> Book.objects.values('publisher').annotate(Count('publisher')).count()
|
|
4
|
|
|
|
>>> Book.objects.annotate(Count('publisher')).values('publisher').count()
|
|
6
|
|
|
|
>>> publishers = Publisher.objects.filter(id__in=(1,2))
|
|
>>> publishers
|
|
[<Publisher: Apress>, <Publisher: Sams>]
|
|
|
|
>>> publishers = publishers.annotate(n_books=models.Count('book'))
|
|
>>> publishers[0].n_books
|
|
2
|
|
|
|
>>> publishers
|
|
[<Publisher: Apress>, <Publisher: Sams>]
|
|
|
|
>>> books = Book.objects.filter(publisher__in=publishers)
|
|
>>> books
|
|
[<Book: Practical Django Projects>, <Book: Sams Teach Yourself Django in 24 Hours>, <Book: The Definitive Guide to Django: Web Development Done Right>]
|
|
|
|
>>> publishers
|
|
[<Publisher: Apress>, <Publisher: Sams>]
|
|
|
|
|
|
# Regression for 10666 - inherited fields work with annotations and aggregations
|
|
>>> HardbackBook.objects.aggregate(n_pages=Sum('book_ptr__pages'))
|
|
{'n_pages': 2078}
|
|
|
|
>>> HardbackBook.objects.aggregate(n_pages=Sum('pages'))
|
|
{'n_pages': 2078}
|
|
|
|
>>> HardbackBook.objects.annotate(n_authors=Count('book_ptr__authors')).values('name','n_authors')
|
|
[{'n_authors': 2, 'name': u'Artificial Intelligence: A Modern Approach'}, {'n_authors': 1, 'name': u'Paradigms of Artificial Intelligence Programming: Case Studies in Common Lisp'}]
|
|
|
|
>>> HardbackBook.objects.annotate(n_authors=Count('authors')).values('name','n_authors')
|
|
[{'n_authors': 2, 'name': u'Artificial Intelligence: A Modern Approach'}, {'n_authors': 1, 'name': u'Paradigms of Artificial Intelligence Programming: Case Studies in Common Lisp'}]
|
|
|
|
# Regression for #10766 - Shouldn't be able to reference an aggregate fields in an an aggregate() call.
|
|
>>> Book.objects.all().annotate(mean_age=Avg('authors__age')).annotate(Avg('mean_age'))
|
|
Traceback (most recent call last):
|
|
...
|
|
FieldError: Cannot compute Avg('mean_age'): 'mean_age' is an aggregate
|
|
|
|
"""
|
|
}
|
|
|
|
def run_stddev_tests():
|
|
"""Check to see if StdDev/Variance tests should be run.
|
|
|
|
Stddev and Variance are not guaranteed to be available for SQLite, and
|
|
are not available for PostgreSQL before 8.2.
|
|
"""
|
|
if settings.DATABASES[DEFAULT_DB_ALIAS]['ENGINE'] == 'django.db.backends.sqlite3':
|
|
return False
|
|
|
|
class StdDevPop(object):
|
|
sql_function = 'STDDEV_POP'
|
|
|
|
try:
|
|
connection.ops.check_aggregate_support(StdDevPop())
|
|
except:
|
|
return False
|
|
return True
|
|
|
|
if run_stddev_tests():
|
|
__test__['API_TESTS'] += """
|
|
>>> Book.objects.aggregate(StdDev('pages'))
|
|
{'pages__stddev': 311.46...}
|
|
|
|
>>> Book.objects.aggregate(StdDev('rating'))
|
|
{'rating__stddev': 0.60...}
|
|
|
|
>>> Book.objects.aggregate(StdDev('price'))
|
|
{'price__stddev': 24.16...}
|
|
|
|
|
|
>>> Book.objects.aggregate(StdDev('pages', sample=True))
|
|
{'pages__stddev': 341.19...}
|
|
|
|
>>> Book.objects.aggregate(StdDev('rating', sample=True))
|
|
{'rating__stddev': 0.66...}
|
|
|
|
>>> Book.objects.aggregate(StdDev('price', sample=True))
|
|
{'price__stddev': 26.46...}
|
|
|
|
|
|
>>> Book.objects.aggregate(Variance('pages'))
|
|
{'pages__variance': 97010.80...}
|
|
|
|
>>> Book.objects.aggregate(Variance('rating'))
|
|
{'rating__variance': 0.36...}
|
|
|
|
>>> Book.objects.aggregate(Variance('price'))
|
|
{'price__variance': 583.77...}
|
|
|
|
|
|
>>> Book.objects.aggregate(Variance('pages', sample=True))
|
|
{'pages__variance': 116412.96...}
|
|
|
|
>>> Book.objects.aggregate(Variance('rating', sample=True))
|
|
{'rating__variance': 0.44...}
|
|
|
|
>>> Book.objects.aggregate(Variance('price', sample=True))
|
|
{'price__variance': 700.53...}
|
|
|
|
"""
|