def deep_prefetch_related_objects(objects, lookups):
"""
Helper function for prefetch_related functionality.
Populates prefetched objects caches for a list of results
from a QuerySet.
Differs from :meth:`django.db.models.query.prefetch_related_objects`
in that that it can prefetch "non-strict" lookups through GFK.
:param objects: result cache of base queryset.
:param lookups: list with lookups.
"""
#How it works
#------------
# - Since data is not of same type - "unit" of data is a model instance
# (object), not QuerySet.
# - Goals that I was tried to achieve:
# - Prefetching 'non-strict' lookups.
# - Absence of doubled queries to DB, other redundancy in processing
# - and infinite recursions.
# - Simplicity of design in comparison with original
# `django.db.models.query.prefetch_related_objects`
# - Sets are used to avoid duplicates.
# - Python's sets distinguish objects by value. But task require
# to distinguish objects by id, not by value, because two objects
# that are same by value, might be included in two different
# querysets, and both of them must be populated with cache.
# - BUT fetched cache data for object are distinguished by value.
# Rationale is to distinguish unique queries and prevent their
# repeated execution. We don't care about "id" of queries, we care
# about their value. Composite identifier for query is
# object and lookup for which query being constructed.
# Only one query exists for one object (by value) and lookup part.
# In function there slightly more complex structure than
# `obj -> lookup` used for storing data for traversed lookups
# (``seen``) - that's done to reduce redundancy in data, but
# "primary key" for stroed data is `obj -> lookup`.
# - Data flows through `buffer`, during processing.
# Objects discovered while traversing DB structure are being added
# and processed objects are removed.
#todo beauty and refactoring
if len(objects) == 0:
return # nothing to do
#lookup -> model -> {(id(obj), obj), ...}
#id(obj) - because objects must be distinguished by id.
#DefaultOrederedDict because order of lookups matter,
#see tests.LookupOrderingTest.test_order of Django test suite.
buffer = DefaultOrderedDict(lambda: defaultdict(set))
update_buffer(buffer, objects, reversed(lookups))
seen = tree() # model -> attr ->
# single -> bool
# cache_name -> str
# cache ->
# obj -> [cache]
while True:
try:
lookup = last(buffer.keys())
except TypeError:
break
try:
model, current = buffer[lookup].popitem()
except KeyError:
del buffer[lookup]
continue
attr = lookup.split(LOOKUP_SEP)[0]
sample = current.pop()
current.add(sample)
_, object = sample
prefetcher, _, attr_found, _ = get_prefetcher(object, attr)
#Lookup is not valid for that object, it must be skipped.
#No exception, because data it is that data is of diffrerent types,
#so - such situation is normal.
if not attr_found:
continue
if LOOKUP_SEP not in lookup and prefetcher is None:
raise ValueError("'%s' does not resolve to a item that supports "
"prefetching - this is an invalid parameter to "
"prefetch_related()." % lookup)
if prefetcher is None:
clipped = clip_lookup(lookup)
if clipped:
update_buffer(
buffer,
filter(is_not_none,
[getattr(o, attr) for _, o in current]), [clipped])
continue
to_discard = set()
for e in current: # no need to query for already prefetched data
obj = e[1]
obj_model = obj.__class__
p, d, _, is_fetched = get_prefetcher(obj, attr)
cache = None
if is_fetched: # case of Django internal cache
single, cache_name = get_info(p, d)
cache = get_cache(obj, single, cache_name, attr)
single, cache_name = get_info(p, d)
update_seen(seen, model, attr, single, cache_name, obj, cache)
to_discard.add(e)
elif (model in seen and # case of `seen`
attr in seen[obj_model] and obj
in seen[obj_model][attr]['cache']):
single = seen[obj_model][attr]['single']
cache_name = seen[obj_model][attr]['cache_name']
cache = seen[obj_model][attr]['cache'][obj]
to_discard.add(e)
set_cache(obj, single, cache, cache_name, attr)
if cache is not None and len(cache) != 0: # if data was cached
clipped = clip_lookup(lookup) # it still must get
if clipped: # into `buffer`.
update_buffer(buffer, cache, [clipped])
current -= to_discard
if current:
prefetch_qs, rel_attr_fn, cur_attr_fn, single, cache_name = \
prefetcher.get_prefetch_queryset(
list(map(itemgetter(1), current))
)
# Prefetch lookups from prefetch queries are merged
# with processed lookups.
additional_lookups = getattr(prefetch_qs,
'_prefetch_related_lookups', [])
if additional_lookups:
setattr(prefetch_qs, '_prefetch_related_lookups', [])
discovered = list(prefetch_qs)
if discovered:
lookups_for_discovered = additional_lookups
clipped_lookup = LOOKUP_SEP.join(lookup.split(LOOKUP_SEP)[1:])
if len(clipped_lookup) > 0:
lookups_for_discovered = chain([clipped_lookup],
additional_lookups)
if lookups_for_discovered:
reversed_lookups = reversed(list(lookups_for_discovered))
update_buffer(buffer, discovered, reversed_lookups)
rel_to_cur = defaultdict(list)
for obj in discovered:
val = rel_attr_fn(obj)
rel_to_cur[val].append(obj)
for pair in current: # queried data is set up to objects
obj = pair[1]
val = cur_attr_fn(obj)
cache = rel_to_cur.get(val, [])
update_seen(seen, model, attr, single, cache_name, obj, cache)
set_cache(obj, single, cache, cache_name, attr)
def test_bug(self):
prefetcher = get_prefetcher(self.rooms[0], 'house')[0]
queryset = prefetcher.get_prefetch_queryset(list(Room.objects.all()))[0]
self.assertNotIn(' JOIN ', force_text(queryset.query))
def test_bug(self):
prefetcher = get_prefetcher(self.rooms[0], 'house')[0]
queryset = prefetcher.get_prefetch_queryset(list(Room.objects.all()))[0]
self.assertNotIn(' JOIN ', force_text(queryset.query))
def deep_prefetch_related_objects(objects, lookups):
"""
Helper function for prefetch_related functionality.
Populates prefetched objects caches for a list of results
from a QuerySet.
Differs from :meth:`django.db.models.query.prefetch_related_objects`
in that that it can prefetch "non-strict" lookups through GFK.
:param objects: result cache of base queryset.
:param lookups: list with lookups.
"""
#How it works
#------------
# - Since data is not of same type - "unit" of data is a model instance
# (object), not QuerySet.
# - Goals that I was tried to achieve:
# - Prefetching 'non-strict' lookups.
# - Absence of doubled queries to DB, other redundancy in processing
# - and infinite recursions.
# - Simplicity of design in comparison with original
# `django.db.models.query.prefetch_related_objects`
# - Sets are used to avoid duplicates.
# - Python's sets distinguish objects by value. But task require
# to distinguish objects by id, not by value, because two objects
# that are same by value, might be included in two different
# querysets, and both of them must be populated with cache.
# - BUT fetched cache data for object are distinguished by value.
# Rationale is to distinguish unique queries and prevent their
# repeated execution. We don't care about "id" of queries, we care
# about their value. Composite identifier for query is
# object and lookup for which query being constructed.
# Only one query exists for one object (by value) and lookup part.
# In function there slightly more complex structure than
# `obj -> lookup` used for storing data for traversed lookups
# (``seen``) - that's done to reduce redundancy in data, but
# "primary key" for stroed data is `obj -> lookup`.
# - Data flows through `buffer`, during processing.
# Objects discovered while traversing DB structure are being added
# and processed objects are removed.
#todo beauty and refactoring
if len(objects) == 0:
return # nothing to do
#lookup -> model -> {(id(obj), obj), ...}
#id(obj) - because objects must be distinguished by id.
#DefaultOrederedDict because order of lookups matter,
#see tests.LookupOrderingTest.test_order of Django test suite.
buffer = DefaultOrderedDict(lambda: defaultdict(set))
update_buffer(buffer, objects, reversed(lookups))
seen = tree() # model -> attr ->
# single -> bool
# cache_name -> str
# cache ->
# obj -> [cache]
while True:
try:
lookup = last(buffer.keys())
except TypeError:
break
try:
model, current = buffer[lookup].popitem()
except KeyError:
del buffer[lookup]
continue
attr = lookup.split(LOOKUP_SEP)[0]
sample = current.pop()
current.add(sample)
_, object = sample
prefetcher, _, attr_found, _ = get_prefetcher(object, attr)
#Lookup is not valid for that object, it must be skipped.
#No exception, because data it is that data is of diffrerent types,
#so - such situation is normal.
if not attr_found:
continue
if LOOKUP_SEP not in lookup and prefetcher is None:
raise ValueError("'%s' does not resolve to a item that supports "
"prefetching - this is an invalid parameter to "
"prefetch_related()." % lookup)
if prefetcher is None:
clipped = clip_lookup(lookup)
if clipped:
update_buffer(
buffer,
filter(is_not_none, [getattr(o, attr)
for _, o in current]),
[clipped])
continue
to_discard = set()
for e in current: # no need to query for already prefetched data
obj = e[1]
obj_model = obj.__class__
p, d, _, is_fetched = get_prefetcher(obj, attr)
cache = None
if is_fetched: # case of Django internal cache
single, cache_name = get_info(p, d)
cache = get_cache(obj, single, cache_name, attr)
single, cache_name = get_info(p, d)
update_seen(seen, model, attr, single, cache_name, obj, cache)
to_discard.add(e)
elif (model in seen and # case of `seen`
attr in seen[obj_model] and
obj in seen[obj_model][attr]['cache']):
single = seen[obj_model][attr]['single']
cache_name = seen[obj_model][attr]['cache_name']
cache = seen[obj_model][attr]['cache'][obj]
to_discard.add(e)
set_cache(obj, single, cache, cache_name, attr)
if cache is not None and len(cache) != 0: # if data was cached
clipped = clip_lookup(lookup) # it still must get
if clipped: # into `buffer`.
update_buffer(buffer, cache, [clipped])
current -= to_discard
if current:
prefetch_qs, rel_attr_fn, cur_attr_fn, single, cache_name = \
prefetcher.get_prefetch_query_set(
map(itemgetter(1), current)
)
#prefetch lookups from prefetch queries are merged into processing.
additional_lookups = getattr(prefetch_qs,
'_prefetch_related_lookups', [])
if additional_lookups:
setattr(prefetch_qs, '_prefetch_related_lookups', [])
discovered = list(prefetch_qs)
lookups_for_discovered = additional_lookups
clipped_lookup = LOOKUP_SEP.join(lookup.split(LOOKUP_SEP)[1:])
if len(clipped_lookup) > 0:
lookups_for_discovered = chain([clipped_lookup],
additional_lookups)
if lookups_for_discovered:
update_buffer(buffer, discovered, reversed(list(lookups_for_discovered)))
rel_to_cur = defaultdict(list)
for obj in discovered:
val = rel_attr_fn(obj)
rel_to_cur[val].append(obj)
for pair in current: # queried data is set up to objects
obj = pair[1]
val = cur_attr_fn(obj)
cache = rel_to_cur.get(val, [])
update_seen(seen, model, attr, single, cache_name, obj, cache)
set_cache(obj, single, cache, cache_name, attr)
