| 1 |
""" |
| 2 |
Interfaces for serializing Django objects. |
| 3 |
|
| 4 |
Usage:: |
| 5 |
|
| 6 |
from django.core import serializers |
| 7 |
json = serializers.serialize("json", some_queryset) |
| 8 |
objects = list(serializers.deserialize("json", json)) |
| 9 |
|
| 10 |
To add your own serializers, use the SERIALIZATION_MODULES setting:: |
| 11 |
|
| 12 |
SERIALIZATION_MODULES = { |
| 13 |
"csv": "path.to.csv.serializer", |
| 14 |
"txt": "path.to.txt.serializer", |
| 15 |
} |
| 16 |
|
| 17 |
""" |
| 18 |
|
| 19 |
import importlib |
| 20 |
|
| 21 |
from django.apps import apps |
| 22 |
from django.conf import settings |
| 23 |
from django.core.serializers.base import SerializerDoesNotExist |
| 24 |
|
| 25 |
# Built-in serializers |
| 26 |
BUILTIN_SERIALIZERS = { |
| 27 |
"xml": "django.core.serializers.xml_serializer", |
| 28 |
"python": "django.core.serializers.python", |
| 29 |
"json": "django.core.serializers.json", |
| 30 |
"yaml": "django.core.serializers.pyyaml", |
| 31 |
"jsonl": "django.core.serializers.jsonl", |
| 32 |
} |
| 33 |
|
| 34 |
_serializers = {} |
| 35 |
|
| 36 |
|
| 37 |
class BadSerializer: |
| 38 |
""" |
| 39 |
Stub serializer to hold exception raised during registration |
| 40 |
|
| 41 |
This allows the serializer registration to cache serializers and if there |
| 42 |
is an error raised in the process of creating a serializer it will be |
| 43 |
raised and passed along to the caller when the serializer is used. |
| 44 |
""" |
| 45 |
|
| 46 |
internal_use_only = False |
| 47 |
|
| 48 |
def __init__(self, exception): |
| 49 |
self.exception = exception |
| 50 |
|
| 51 |
def __call__(self, *args, **kwargs): |
| 52 |
raise self.exception |
| 53 |
|
| 54 |
|
| 55 |
def register_serializer(format, serializer_module, serializers=None): |
| 56 |
"""Register a new serializer. |
| 57 |
|
| 58 |
``serializer_module`` should be the fully qualified module name |
| 59 |
for the serializer. |
| 60 |
|
| 61 |
If ``serializers`` is provided, the registration will be added |
| 62 |
to the provided dictionary. |
| 63 |
|
| 64 |
If ``serializers`` is not provided, the registration will be made |
| 65 |
directly into the global register of serializers. Adding serializers |
| 66 |
directly is not a thread-safe operation. |
| 67 |
""" |
| 68 |
if serializers is None and not _serializers: |
| 69 |
_load_serializers() |
| 70 |
|
| 71 |
try: |
| 72 |
module = importlib.import_module(serializer_module) |
| 73 |
except ImportError as exc: |
| 74 |
bad_serializer = BadSerializer(exc) |
| 75 |
|
| 76 |
module = type( |
| 77 |
"BadSerializerModule", |
| 78 |
(), |
| 79 |
{ |
| 80 |
"Deserializer": bad_serializer, |
| 81 |
"Serializer": bad_serializer, |
| 82 |
}, |
| 83 |
) |
| 84 |
|
| 85 |
if serializers is None: |
| 86 |
_serializers[format] = module |
| 87 |
else: |
| 88 |
serializers[format] = module |
| 89 |
|
| 90 |
|
| 91 |
def unregister_serializer(format): |
| 92 |
"Unregister a given serializer. This is not a thread-safe operation." |
| 93 |
if not _serializers: |
| 94 |
_load_serializers() |
| 95 |
if format not in _serializers: |
| 96 |
raise SerializerDoesNotExist(format) |
| 97 |
del _serializers[format] |
| 98 |
|
| 99 |
|
| 100 |
def get_serializer(format): |
| 101 |
if not _serializers: |
| 102 |
_load_serializers() |
| 103 |
if format not in _serializers: |
| 104 |
raise SerializerDoesNotExist(format) |
| 105 |
return _serializers[format].Serializer |
| 106 |
|
| 107 |
|
| 108 |
def get_serializer_formats(): |
| 109 |
if not _serializers: |
| 110 |
_load_serializers() |
| 111 |
return list(_serializers) |
| 112 |
|
| 113 |
|
| 114 |
def get_public_serializer_formats(): |
| 115 |
if not _serializers: |
| 116 |
_load_serializers() |
| 117 |
return [k for k, v in _serializers.items() if not v.Serializer.internal_use_only] |
| 118 |
|
| 119 |
|
| 120 |
def get_deserializer(format): |
| 121 |
if not _serializers: |
| 122 |
_load_serializers() |
| 123 |
if format not in _serializers: |
| 124 |
raise SerializerDoesNotExist(format) |
| 125 |
return _serializers[format].Deserializer |
| 126 |
|
| 127 |
|
| 128 |
def serialize(format, queryset, **options): |
| 129 |
""" |
| 130 |
Serialize a queryset (or any iterator that returns database objects) using |
| 131 |
a certain serializer. |
| 132 |
""" |
| 133 |
s = get_serializer(format)() |
| 134 |
s.serialize(queryset, **options) |
| 135 |
return s.getvalue() |
| 136 |
|
| 137 |
|
| 138 |
def deserialize(format, stream_or_string, **options): |
| 139 |
""" |
| 140 |
Deserialize a stream or a string. Return an iterator that yields ``(obj, |
| 141 |
m2m_relation_dict)``, where ``obj`` is an instantiated -- but *unsaved* -- |
| 142 |
object, and ``m2m_relation_dict`` is a dictionary of ``{m2m_field_name : |
| 143 |
list_of_related_objects}``. |
| 144 |
""" |
| 145 |
d = get_deserializer(format) |
| 146 |
return d(stream_or_string, **options) |
| 147 |
|
| 148 |
|
| 149 |
def _load_serializers(): |
| 150 |
""" |
| 151 |
Register built-in and settings-defined serializers. This is done lazily so |
| 152 |
that user code has a chance to (e.g.) set up custom settings without |
| 153 |
needing to be careful of import order. |
| 154 |
""" |
| 155 |
global _serializers |
| 156 |
serializers = {} |
| 157 |
for format in BUILTIN_SERIALIZERS: |
| 158 |
register_serializer(format, BUILTIN_SERIALIZERS[format], serializers) |
| 159 |
if hasattr(settings, "SERIALIZATION_MODULES"): |
| 160 |
for format in settings.SERIALIZATION_MODULES: |
| 161 |
register_serializer( |
| 162 |
format, settings.SERIALIZATION_MODULES[format], serializers |
| 163 |
) |
| 164 |
_serializers = serializers |
| 165 |
|
| 166 |
|
| 167 |
def sort_dependencies(app_list, allow_cycles=False): |
| 168 |
"""Sort a list of (app_config, models) pairs into a single list of models. |
| 169 |
|
| 170 |
The single list of models is sorted so that any model with a natural key |
| 171 |
is serialized before a normal model, and any model with a natural key |
| 172 |
dependency has it's dependencies serialized first. |
| 173 |
|
| 174 |
If allow_cycles is True, return the best-effort ordering that will respect |
| 175 |
most of dependencies but ignore some of them to break the cycles. |
| 176 |
""" |
| 177 |
# Process the list of models, and get the list of dependencies |
| 178 |
model_dependencies = [] |
| 179 |
models = set() |
| 180 |
for app_config, model_list in app_list: |
| 181 |
if model_list is None: |
| 182 |
model_list = app_config.get_models() |
| 183 |
|
| 184 |
for model in model_list: |
| 185 |
models.add(model) |
| 186 |
# Add any explicitly defined dependencies |
| 187 |
if hasattr(model, "natural_key"): |
| 188 |
deps = getattr(model.natural_key, "dependencies", []) |
| 189 |
if deps: |
| 190 |
deps = [apps.get_model(dep) for dep in deps] |
| 191 |
else: |
| 192 |
deps = [] |
| 193 |
|
| 194 |
# Now add a dependency for any FK relation with a model that |
| 195 |
# defines a natural key |
| 196 |
for field in model._meta.fields: |
| 197 |
if field.remote_field: |
| 198 |
rel_model = field.remote_field.model |
| 199 |
if hasattr(rel_model, "natural_key") and rel_model != model: |
| 200 |
deps.append(rel_model) |
| 201 |
# Also add a dependency for any simple M2M relation with a model |
| 202 |
# that defines a natural key. M2M relations with explicit through |
| 203 |
# models don't count as dependencies. |
| 204 |
for field in model._meta.many_to_many: |
| 205 |
if field.remote_field.through._meta.auto_created: |
| 206 |
rel_model = field.remote_field.model |
| 207 |
if hasattr(rel_model, "natural_key") and rel_model != model: |
| 208 |
deps.append(rel_model) |
| 209 |
model_dependencies.append((model, deps)) |
| 210 |
|
| 211 |
model_dependencies.reverse() |
| 212 |
# Now sort the models to ensure that dependencies are met. This |
| 213 |
# is done by repeatedly iterating over the input list of models. |
| 214 |
# If all the dependencies of a given model are in the final list, |
| 215 |
# that model is promoted to the end of the final list. This process |
| 216 |
# continues until the input list is empty, or we do a full iteration |
| 217 |
# over the input models without promoting a model to the final list. |
| 218 |
# If we do a full iteration without a promotion, that means there are |
| 219 |
# circular dependencies in the list. |
| 220 |
model_list = [] |
| 221 |
while model_dependencies: |
| 222 |
skipped = [] |
| 223 |
changed = False |
| 224 |
while model_dependencies: |
| 225 |
model, deps = model_dependencies.pop() |
| 226 |
|
| 227 |
# If all of the models in the dependency list are either already |
| 228 |
# on the final model list, or not on the original serialization list, |
| 229 |
# then we've found another model with all it's dependencies satisfied. |
| 230 |
if all(d not in models or d in model_list for d in deps): |
| 231 |
model_list.append(model) |
| 232 |
changed = True |
| 233 |
else: |
| 234 |
skipped.append((model, deps)) |
| 235 |
if not changed: |
| 236 |
if allow_cycles: |
| 237 |
# If cycles are allowed, add the last skipped model and ignore |
| 238 |
# its dependencies. This could be improved by some graph |
| 239 |
# analysis to ignore as few dependencies as possible. |
| 240 |
model, _ = skipped.pop() |
| 241 |
model_list.append(model) |
| 242 |
else: |
| 243 |
raise RuntimeError( |
| 244 |
"Can't resolve dependencies for %s in serialized app list." |
| 245 |
% ", ".join( |
| 246 |
model._meta.label |
| 247 |
for model, deps in sorted( |
| 248 |
skipped, key=lambda obj: obj[0].__name__ |
| 249 |
) |
| 250 |
), |
| 251 |
) |
| 252 |
model_dependencies = skipped |
| 253 |
|
| 254 |
return model_list |