efro.dataclassio

34class Codec(Enum):
35    """Specifies expected data format exported to or imported from."""
36
37    # Use only types that will translate cleanly to/from json: lists,
38    # dicts with str keys, bools, ints, floats, and None.
39    JSON = 'json'
40
41    # Mostly like JSON but passes bytes and datetime objects through
42    # as-is instead of converting them to json-friendly types.
43    FIRESTORE = 'firestore'

 61class DataclassFieldLookup(Generic[T]):
 62    """Get info about nested dataclass fields in type-safe way."""
 63
 64    def __init__(self, cls: type[T]) -> None:
 65        self.cls = cls
 66
 67    def path(self, callback: Callable[[T], Any]) -> str:
 68        """Look up a path on child dataclass fields.
 69
 70        example:
 71          DataclassFieldLookup(MyType).path(lambda obj: obj.foo.bar)
 72
 73        The above example will return the string 'foo.bar' or something
 74        like 'f.b' if the dataclasses have custom storage names set.
 75        It will also be static-type-checked, triggering an error if
 76        MyType.foo.bar is not a valid path. Note, however, that the
 77        callback technically allows any return value but only nested
 78        dataclasses and their fields will succeed.
 79        """
 80
 81        # We tell the type system that we are returning an instance
 82        # of our class, which allows it to perform type checking on
 83        # member lookups. In reality, however, we are providing a
 84        # special object which captures path lookups, so we can build
 85        # a string from them.
 86        if not TYPE_CHECKING:
 87            out = callback(_PathCapture(self.cls))
 88            if not isinstance(out, _PathCapture):
 89                raise TypeError(
 90                    f'Expected a valid path under'
 91                    f' the provided object; got a {type(out)}.'
 92                )
 93            return out.path
 94        return ''
 95
 96    def paths(self, callback: Callable[[T], list[Any]]) -> list[str]:
 97        """Look up multiple paths on child dataclass fields.
 98
 99        Functionality is identical to path() but for multiple paths at once.
100
101        example:
102          DataclassFieldLookup(MyType).paths(lambda obj: [obj.foo, obj.bar])
103        """
104        outvals: list[str] = []
105        if not TYPE_CHECKING:
106            outs = callback(_PathCapture(self.cls))
107            assert isinstance(outs, list)
108            for out in outs:
109                if not isinstance(out, _PathCapture):
110                    raise TypeError(
111                        f'Expected a valid path under'
112                        f' the provided object; got a {type(out)}.'
113                    )
114                outvals.append(out.path)
115        return outvals

156class IOAttrs:
157    """For specifying io behavior in annotations.
158
159    'storagename', if passed, is the name used when storing to json/etc.
160
161    'store_default' can be set to False to avoid writing values when
162        equal to the default value. Note that this requires the
163        dataclass field to define a default or default_factory or for
164        its IOAttrs to define a soft_default value.
165
166    'whole_days', if True, requires datetime values to be exactly on day
167        boundaries (see efro.util.utc_today()).
168
169    'whole_hours', if True, requires datetime values to lie exactly on
170        hour boundaries (see efro.util.utc_this_hour()).
171
172    'whole_minutes', if True, requires datetime values to lie exactly on
173        minute boundaries (see efro.util.utc_this_minute()).
174
175    'soft_default', if passed, injects a default value into dataclass
176        instantiation when the field is not present in the input data.
177        This allows dataclasses to add new non-optional fields while
178        gracefully 'upgrading' old data. Note that when a soft_default
179        is present it will take precedence over field defaults when
180        determining whether to store a value for a field with
181        store_default=False (since the soft_default value is what we'll
182        get when reading that same data back in when the field is
183        omitted).
184
185    'soft_default_factory' is similar to 'default_factory' in dataclass
186        fields; it should be used instead of 'soft_default' for mutable
187        types such as lists to prevent a single default object from
188        unintentionally changing over time.
189
190    'enum_fallback', if provided, specifies an enum value that can be
191        substituted in the case of unrecognized input values. This can
192        allow newer data to remain loadable in older environments. Note
193        that 'lossy' must be enabled in the top level load call for this
194        to apply, since it can fundamentally modify data.
195    """
196
197    # A sentinel object to detect if a parameter is supplied or not. Use
198    # a class to give it a better repr.
199    class _MissingType:
200        pass
201
202    MISSING = _MissingType()
203
204    storagename: str | None = None
205    store_default: bool = True
206    whole_days: bool = False
207    whole_hours: bool = False
208    whole_minutes: bool = False
209    soft_default: Any = MISSING
210    soft_default_factory: Callable[[], Any] | _MissingType = MISSING
211    enum_fallback: Enum | None = None
212
213    def __init__(
214        self,
215        storagename: str | None = storagename,
216        *,
217        store_default: bool = store_default,
218        whole_days: bool = whole_days,
219        whole_hours: bool = whole_hours,
220        whole_minutes: bool = whole_minutes,
221        soft_default: Any = MISSING,
222        soft_default_factory: Callable[[], Any] | _MissingType = MISSING,
223        enum_fallback: Enum | None = None,
224    ):
225        # Only store values that differ from class defaults to keep
226        # our instances nice and lean.
227        cls = type(self)
228        if storagename != cls.storagename:
229            self.storagename = storagename
230        if store_default != cls.store_default:
231            self.store_default = store_default
232        if whole_days != cls.whole_days:
233            self.whole_days = whole_days
234        if whole_hours != cls.whole_hours:
235            self.whole_hours = whole_hours
236        if whole_minutes != cls.whole_minutes:
237            self.whole_minutes = whole_minutes
238        if soft_default is not cls.soft_default:
239            # Do what dataclasses does with its default types and
240            # tell the user to use factory for mutable ones.
241            if isinstance(soft_default, (list, dict, set)):
242                raise ValueError(
243                    f'mutable {type(soft_default)} is not allowed'
244                    f' for soft_default; use soft_default_factory.'
245                )
246            self.soft_default = soft_default
247        if soft_default_factory is not cls.soft_default_factory:
248            self.soft_default_factory = soft_default_factory
249            if self.soft_default is not cls.soft_default:
250                raise ValueError(
251                    'Cannot set both soft_default and soft_default_factory'
252                )
253        if enum_fallback is not cls.enum_fallback:
254            self.enum_fallback = enum_fallback
255
256    def validate_for_field(self, cls: type, field: dataclasses.Field) -> None:
257        """Ensure the IOAttrs instance is ok to use with the provided field."""
258
259        # Turning off store_default requires the field to have either
260        # a default or a default_factory or for us to have soft equivalents.
261
262        if not self.store_default:
263            field_default_factory: Any = field.default_factory
264            if (
265                field_default_factory is dataclasses.MISSING
266                and field.default is dataclasses.MISSING
267                and self.soft_default is self.MISSING
268                and self.soft_default_factory is self.MISSING
269            ):
270                raise TypeError(
271                    f'Field {field.name} of {cls} has'
272                    f' neither a default nor a default_factory'
273                    f' and IOAttrs contains neither a soft_default'
274                    f' nor a soft_default_factory;'
275                    f' store_default=False cannot be set for it.'
276                )
277
278    def validate_datetime(
279        self, value: datetime.datetime, fieldpath: str
280    ) -> None:
281        """Ensure a datetime value meets our value requirements."""
282        if self.whole_days:
283            if any(
284                x != 0
285                for x in (
286                    value.hour,
287                    value.minute,
288                    value.second,
289                    value.microsecond,
290                )
291            ):
292                raise ValueError(
293                    f'Value {value} at {fieldpath} is not a whole day.'
294                )
295        elif self.whole_hours:
296            if any(
297                x != 0 for x in (value.minute, value.second, value.microsecond)
298            ):
299                raise ValueError(
300                    f'Value {value} at {fieldpath}' f' is not a whole hour.'
301                )
302        elif self.whole_minutes:
303            if any(x != 0 for x in (value.second, value.microsecond)):
304                raise ValueError(
305                    f'Value {value} at {fieldpath}' f' is not a whole minute.'
306                )

46class IOExtendedData:
47    """A class that data types can inherit from for extra functionality."""
48
49    def will_output(self) -> None:
50        """Called before data is sent to an outputter.
51
52        Can be overridden to validate or filter data before
53        sending it on its way.
54        """
55
56    @classmethod
57    def will_input(cls, data: dict) -> None:
58        """Called on raw data before a class instance is created from it.
59
60        Can be overridden to migrate old data formats to new, etc.
61        """
62
63    def did_input(self) -> None:
64        """Called on a class instance after created from data.
65
66        Can be useful to correct values from the db, etc. in the
67        type-safe form.
68        """
69
70    # pylint: disable=useless-return
71
72    @classmethod
73    def handle_input_error(cls, exc: Exception) -> Self | None:
74        """Called when an error occurs during input decoding.
75
76        This allows a type to optionally return substitute data
77        to be used in place of the failed decode. If it returns
78        None, the original exception is re-raised.
79
80        It is generally a bad idea to apply catch-alls such as this,
81        as it can lead to silent data loss. This should only be used
82        in specific cases such as user settings where an occasional
83        reset is harmless and is preferable to keeping all contained
84        enums and other values backward compatible indefinitely.
85        """
86        del exc  # Unused.
87
88        # By default we let things fail.
89        return None
90
91    # pylint: enable=useless-return

 97class IOMultiType(Generic[EnumT]):
 98    """A base class for types that can map to multiple dataclass types.
 99
100    This enables usage of high level base classes (for example
101    a 'Message' type) in annotations, with dataclassio automatically
102    serializing & deserializing dataclass subclasses based on their
103    type ('MessagePing', 'MessageChat', etc.)
104
105    Standard usage involves creating a class which inherits from this
106    one which acts as a 'registry', and then creating dataclass classes
107    inheriting from that registry class. Dataclassio will then do the
108    right thing when that registry class is used in type annotations.
109
110    See tests/test_efro/test_dataclassio.py for examples.
111    """
112
113    @classmethod
114    def get_type(cls, type_id: EnumT) -> type[Self]:
115        """Return a specific subclass given a type-id."""
116        raise NotImplementedError()
117
118    @classmethod
119    def get_type_id(cls) -> EnumT:
120        """Return the type-id for this subclass."""
121        raise NotImplementedError()
122
123    @classmethod
124    def get_type_id_type(cls) -> type[EnumT]:
125        """Return the Enum type this class uses as its type-id."""
126        out: type[EnumT] = cls.__orig_bases__[0].__args__[0]  # type: ignore
127        assert issubclass(out, Enum)
128        return out
129
130    @classmethod
131    def get_type_id_storage_name(cls) -> str:
132        """Return the key used to store type id in serialized data.
133
134        The default is an obscure value so that it does not conflict
135        with members of individual type attrs, but in some cases one
136        might prefer to serialize it to something simpler like 'type' by
137        overriding this call. One just needs to make sure that no
138        encompassed types serialize anything to 'type' themself.
139        """
140        return '_dciotype'
141
142    # NOTE: Currently (Jan 2025) mypy complains if overrides annotate
143    # return type of 'Self | None'. Substituting their own explicit type
144    # works though (see test_dataclassio).
145    @classmethod
146    def get_unknown_type_fallback(cls) -> Self | None:
147        """Return a fallback object in cases of unrecognized types.
148
149        This can allow newer data to remain readable in older
150        environments. Use caution with this option, however, as it
151        effectively modifies data.
152        """
153        return None

28class JsonStyle(Enum):
29    """Different style types for json."""
30
31    # Single line, no spaces, no sorting. Not deterministic.
32    # Use this where speed is more important than determinism.
33    FAST = 'fast'
34
35    # Single line, no spaces, sorted keys. Deterministic.
36    # Use this when output may be hashed or compared for equality.
37    SORTED = 'sorted'
38
39    # Multiple lines, spaces, sorted keys. Deterministic.
40    # Use this for pretty human readable output.
41    PRETTY = 'pretty'

100def dataclass_from_dict(
101    cls: type[T],
102    values: dict,
103    *,
104    codec: Codec = Codec.JSON,
105    coerce_to_float: bool = True,
106    allow_unknown_attrs: bool = True,
107    discard_unknown_attrs: bool = False,
108    lossy: bool = False,
109) -> T:
110    """Given a dict, return a dataclass of a given type.
111
112    The dict must be formatted to match the specified codec (generally
113    json-friendly object types). This means that sequence values such as
114    tuples or sets should be passed as lists, enums should be passed as
115    their associated values, nested dataclasses should be passed as
116    dicts, etc.
117
118    All values are checked to ensure their types/values are valid.
119
120    Data for attributes of type Any will be checked to ensure they match
121    types supported directly by json. This does not include types such
122    as tuples which are implicitly translated by Python's json module
123    (as this would break the ability to do a lossless round-trip with
124    data).
125
126    If `coerce_to_float` is True, int values passed for float typed
127    fields will be converted to float values. Otherwise, a TypeError is
128    raised.
129
130    If 'allow_unknown_attrs' is False, AttributeErrors will be raised
131    for attributes present in the dict but not on the data class.
132    Otherwise, they will be preserved as part of the instance and
133    included if it is exported back to a dict, unless
134    `discard_unknown_attrs` is True, in which case they will simply be
135    discarded.
136
137    If `lossy` is True, Enum attrs and IOMultiType types are allowed to
138    use any fallbacks defined for them. This can allow older schemas to
139    successfully load newer data, but this can fundamentally modify the
140    data, so the resulting object is flagged as 'lossy' and prevented
141    from being serialized back out by default.
142    """
143    val = _Inputter(
144        cls,
145        codec=codec,
146        coerce_to_float=coerce_to_float,
147        allow_unknown_attrs=allow_unknown_attrs,
148        discard_unknown_attrs=discard_unknown_attrs,
149        lossy=lossy,
150    ).run(values)
151    assert isinstance(val, cls)
152    return val

155def dataclass_from_json(
156    cls: type[T],
157    json_str: str,
158    *,
159    coerce_to_float: bool = True,
160    allow_unknown_attrs: bool = True,
161    discard_unknown_attrs: bool = False,
162    lossy: bool = False,
163) -> T:
164    """Return a dataclass instance given a json string.
165
166    Basically dataclass_from_dict(json.loads(...))
167    """
168
169    return dataclass_from_dict(
170        cls=cls,
171        values=json.loads(json_str),
172        coerce_to_float=coerce_to_float,
173        allow_unknown_attrs=allow_unknown_attrs,
174        discard_unknown_attrs=discard_unknown_attrs,
175        lossy=lossy,
176    )

44def dataclass_to_dict(
45    obj: Any,
46    codec: Codec = Codec.JSON,
47    coerce_to_float: bool = True,
48    discard_extra_attrs: bool = False,
49) -> dict:
50    """Given a dataclass object, return a json-friendly dict.
51
52    All values will be checked to ensure they match the types specified
53    on fields. Note that a limited set of types and data configurations is
54    supported.
55
56    Values with type Any will be checked to ensure they match types supported
57    directly by json. This does not include types such as tuples which are
58    implicitly translated by Python's json module (as this would break
59    the ability to do a lossless round-trip with data).
60
61    If coerce_to_float is True, integer values present on float typed fields
62    will be converted to float in the dict output. If False, a TypeError
63    will be triggered.
64    """
65
66    out = _Outputter(
67        obj,
68        create=True,
69        codec=codec,
70        coerce_to_float=coerce_to_float,
71        discard_extra_attrs=discard_extra_attrs,
72    ).run()
73    assert isinstance(out, dict)
74    return out

77def dataclass_to_json(
78    obj: Any,
79    coerce_to_float: bool = True,
80    pretty: bool = False,
81    sort_keys: bool | None = None,
82) -> str:
83    """Utility function; return a json string from a dataclass instance.
84
85    Basically json.dumps(dataclass_to_dict(...)).
86    By default, keys are sorted for pretty output and not otherwise, but
87    this can be overridden by supplying a value for the 'sort_keys' arg.
88    """
89
90    jdict = dataclass_to_dict(
91        obj=obj, coerce_to_float=coerce_to_float, codec=Codec.JSON
92    )
93    if sort_keys is None:
94        sort_keys = pretty
95    if pretty:
96        return json.dumps(jdict, indent=2, sort_keys=sort_keys)
97    return json.dumps(jdict, separators=(',', ':'), sort_keys=sort_keys)

179def dataclass_validate(
180    obj: Any,
181    coerce_to_float: bool = True,
182    codec: Codec = Codec.JSON,
183    discard_extra_attrs: bool = False,
184) -> None:
185    """Ensure that values in a dataclass instance are the correct types."""
186
187    # Simply run an output pass but tell it not to generate data;
188    # only run validation.
189    _Outputter(
190        obj,
191        create=False,
192        codec=codec,
193        coerce_to_float=coerce_to_float,
194        discard_extra_attrs=discard_extra_attrs,
195    ).run()

198def dataclass_hash(obj: Any, coerce_to_float: bool = True) -> str:
199    """Calculate a hash for the provided dataclass.
200
201    Basically this emits json for the dataclass (with keys sorted
202    to keep things deterministic) and hashes the resulting string.
203    """
204    import hashlib
205    from base64 import urlsafe_b64encode
206
207    json_dict = dataclass_to_dict(
208        obj, codec=Codec.JSON, coerce_to_float=coerce_to_float
209    )
210
211    # Need to sort keys to keep things deterministic.
212    json_str = json.dumps(json_dict, separators=(',', ':'), sort_keys=True)
213
214    sha = hashlib.sha256()
215    sha.update(json_str.encode())
216
217    # Go with urlsafe base64 instead of the usual hex to save some
218    # space, and kill those ugly padding chars at the end.
219    return urlsafe_b64encode(sha.digest()).decode().strip('=')

47def ioprep(cls: type, globalns: dict | None = None) -> None:
48    """Prep a dataclass type for use with this module's functionality.
49
50    Prepping ensures that all types contained in a data class as well as
51    the usage of said types are supported by this module and pre-builds
52    necessary constructs needed for encoding/decoding/etc.
53
54    Prepping will happen on-the-fly as needed, but a warning will be
55    emitted in such cases, as it is better to explicitly prep all used types
56    early in a process to ensure any invalid types or configuration are caught
57    immediately.
58
59    Prepping a dataclass involves evaluating its type annotations, which,
60    as of PEP 563, are stored simply as strings. This evaluation is done
61    with localns set to the class dict (so that types defined in the class
62    can be used) and globalns set to the containing module's class.
63    It is possible to override globalns for special cases such as when
64    prepping happens as part of an execed string instead of within a
65    module.
66    """
67    PrepSession(explicit=True, globalns=globalns).prep_dataclass(
68        cls, recursion_level=0
69    )

72def ioprepped(cls: type[T]) -> type[T]:
73    """Class decorator for easily prepping a dataclass at definition time.
74
75    Note that in some cases it may not be possible to prep a dataclass
76    immediately (such as when its type annotations refer to forward-declared
77    types). In these cases, dataclass_prep() should be explicitly called for
78    the class as soon as possible; ideally at module import time to expose any
79    errors as early as possible in execution.
80    """
81    ioprep(cls)
82    return cls

102def is_ioprepped_dataclass(obj: Any) -> bool:
103    """Return whether the obj is an ioprepped dataclass type or instance."""
104    cls = obj if isinstance(obj, type) else type(obj)
105    return dataclasses.is_dataclass(cls) and hasattr(cls, PREP_ATTR)

85def will_ioprep(cls: type[T]) -> type[T]:
86    """Class decorator hinting that we will prep a class later.
87
88    In some cases (such as recursive types) we cannot use the @ioprepped
89    decorator and must instead call ioprep() explicitly later. However,
90    some of our custom pylint checking behaves differently when the
91    @ioprepped decorator is present, in that case requiring type annotations
92    to be present and not simply forward declared under an "if TYPE_CHECKING"
93    block. (since they are used at runtime).
94
95    The @will_ioprep decorator triggers the same pylint behavior
96    differences as @ioprepped (which are necessary for the later ioprep() call
97    to work correctly) but without actually running any prep itself.
98    """
99    return cls