#----------------------------------------------------------------------------- # Copyright (c) 2012 - 2023, Anaconda, Inc., and Bokeh Contributors. # All rights reserved. # # The full license is in the file LICENSE.txt, distributed with this software. #----------------------------------------------------------------------------- ''' Provide a base class for objects that can have declarative, typed, serializable properties. .. note:: These classes form part of the very low-level machinery that implements the Bokeh model and property system. It is unlikely that any of these classes or their methods will be applicable to any standard usage or to anyone who is not directly developing on Bokeh's own infrastructure. ''' #----------------------------------------------------------------------------- # Boilerplate #----------------------------------------------------------------------------- from __future__ import annotations import logging # isort:skip log = logging.getLogger(__name__) #----------------------------------------------------------------------------- # Imports #----------------------------------------------------------------------------- # Standard library imports import difflib from typing import ( TYPE_CHECKING, Any, Callable, ClassVar, Iterable, Literal, NoReturn, TypedDict, TypeVar, Union, overload, ) from weakref import WeakSet if TYPE_CHECKING: F = TypeVar("F", bound=Callable[..., Any]) def lru_cache(arg: int | None) -> Callable[[F], F]: ... else: from functools import lru_cache # Bokeh imports from ..util.strings import append_docstring, nice_join from ..util.warnings import warn from .property.descriptor_factory import PropertyDescriptorFactory from .property.descriptors import PropertyDescriptor, UnsetValueError from .property.override import Override from .property.singletons import Intrinsic, Undefined from .property.wrappers import PropertyValueContainer from .serialization import ( ObjectRep, Ref, Serializable, Serializer, ) from .types import ID if TYPE_CHECKING: from typing_extensions import NotRequired, TypeAlias from ..client.session import ClientSession from ..server.session import ServerSession from .property.bases import Property from .property.dataspec import DataSpec #----------------------------------------------------------------------------- # Globals and constants #----------------------------------------------------------------------------- __all__ = ( 'abstract', 'HasProps', 'MetaHasProps', 'NonQualified', 'Qualified', ) #----------------------------------------------------------------------------- # General API #----------------------------------------------------------------------------- #----------------------------------------------------------------------------- # Dev API #----------------------------------------------------------------------------- if TYPE_CHECKING: Setter: TypeAlias = Union[ClientSession, ServerSession] C = TypeVar("C", bound=type["HasProps"]) _abstract_classes: WeakSet[type[HasProps]] = WeakSet() def abstract(cls: C) -> C: ''' A decorator to mark abstract base classes derived from |HasProps|. ''' if not issubclass(cls, HasProps): raise TypeError(f"{cls.__name__} is not a subclass of HasProps") _abstract_classes.add(cls) cls.__doc__ = append_docstring(cls.__doc__, _ABSTRACT_ADMONITION) return cls def is_abstract(cls: type[HasProps]) -> bool: return cls in _abstract_classes def is_DataModel(cls: type[HasProps]) -> bool: from ..model import DataModel return issubclass(cls, HasProps) and getattr(cls, "__data_model__", False) and cls != DataModel def _overridden_defaults(class_dict: dict[str, Any]) -> dict[str, Any]: overridden_defaults: dict[str, Any] = {} for name, prop in tuple(class_dict.items()): if isinstance(prop, Override): del class_dict[name] if prop.default_overridden: overridden_defaults[name] = prop.default return overridden_defaults def _generators(class_dict: dict[str, Any]): generators: dict[str, PropertyDescriptorFactory[Any]] = {} for name, generator in tuple(class_dict.items()): if isinstance(generator, PropertyDescriptorFactory): del class_dict[name] generators[name] = generator return generators class _ModelResolver: """ """ _known_models: dict[str, type[HasProps]] def __init__(self) -> None: self._known_models = {} def add(self, cls: type[HasProps]) -> None: if not (issubclass(cls, Local) or cls.__name__.startswith("_")): # update the mapping of view model names to classes, checking for any duplicates previous = self._known_models.get(cls.__qualified_model__, None) if previous is not None and not hasattr(cls, "__implementation__"): raise Warning(f"Duplicate qualified model declaration of '{cls.__qualified_model__}'. Previous definition: {previous}") self._known_models[cls.__qualified_model__] = cls def remove(self, cls: type[HasProps]) -> None: del self._known_models[cls.__qualified_model__] @property def known_models(self) -> dict[str, type[HasProps]]: return dict(self._known_models) def clear_extensions(self) -> None: def is_extension(obj: type[HasProps]) -> bool: return getattr(obj, "__implementation__", None) is not None or \ getattr(obj, "__javascript__", None) is not None or \ getattr(obj, "__css__", None) is not None self._known_models = {key: val for key, val in self._known_models.items() if not is_extension(val)} _default_resolver = _ModelResolver() class MetaHasProps(type): ''' Specialize the construction of |HasProps| classes. This class is a `metaclass`_ for |HasProps| that is responsible for creating and adding the ``PropertyDescriptor`` instances that delegate validation and serialization to |Property| attributes. .. _metaclass: https://docs.python.org/3/reference/datamodel.html#metaclasses ''' __properties__: dict[str, Property[Any]] __overridden_defaults__: dict[str, Any] __themed_values__: dict[str, Any] def __new__(cls, class_name: str, bases: tuple[type, ...], class_dict: dict[str, Any]): ''' ''' overridden_defaults = _overridden_defaults(class_dict) generators = _generators(class_dict) properties = {} for name, generator in generators.items(): descriptors = generator.make_descriptors(name) for descriptor in descriptors: name = descriptor.name if name in class_dict: raise RuntimeError(f"Two property generators both created {class_name}.{name}") class_dict[name] = descriptor properties[name] = descriptor.property class_dict["__properties__"] = properties class_dict["__overridden_defaults__"] = overridden_defaults return super().__new__(cls, class_name, bases, class_dict) def __init__(cls, class_name: str, bases: tuple[type, ...], _) -> None: # HasProps itself may not have any properties defined if class_name == "HasProps": return # Check for improperly redeclared a Property attribute. base_properties: dict[str, Any] = {} for base in (x for x in bases if issubclass(x, HasProps)): base_properties.update(base.properties(_with_props=True)) own_properties = {k: v for k, v in cls.__dict__.items() if isinstance(v, PropertyDescriptor)} redeclared = own_properties.keys() & base_properties.keys() if redeclared: warn(f"Properties {redeclared!r} in class {cls.__name__} were previously declared on a parent " "class. It never makes sense to do this. Redundant properties should be deleted here, or on " "the parent class. Override() can be used to change a default value of a base class property.", RuntimeWarning) # Check for no-op Overrides unused_overrides = cls.__overridden_defaults__.keys() - cls.properties(_with_props=True).keys() if unused_overrides: warn(f"Overrides of {unused_overrides} in class {cls.__name__} does not override anything.", RuntimeWarning) @property def model_class_reverse_map(cls) -> dict[str, type[HasProps]]: return _default_resolver.known_models class Local: """Don't register this class in model registry. """ class Qualified: """Resolve this class by a fully qualified name. """ class NonQualified: """Resolve this class by a non-qualified name. """ class HasProps(Serializable, metaclass=MetaHasProps): ''' Base class for all class types that have Bokeh properties. ''' _initialized: bool = False _property_values: dict[str, Any] _unstable_default_values: dict[str, Any] _unstable_themed_values: dict[str, Any] __view_model__: ClassVar[str] __view_module__: ClassVar[str] __qualified_model__: ClassVar[str] __implementation__: ClassVar[Any] # TODO: specific type __data_model__: ClassVar[bool] @classmethod def __init_subclass__(cls): super().__init_subclass__() # use an explicitly provided view model name if there is one if "__view_model__" not in cls.__dict__: cls.__view_model__ = cls.__qualname__.replace(".", "") if "__view_module__" not in cls.__dict__: cls.__view_module__ = cls.__module__ if "__qualified_model__" not in cls.__dict__: def qualified(): module = cls.__view_module__ model = cls.__view_model__ if issubclass(cls, NonQualified): return model if not issubclass(cls, Qualified): head = module.split(".")[0] if head == "bokeh" or head == "__main__" or "__implementation__" in cls.__dict__: return model return f"{module}.{model}" cls.__qualified_model__ = qualified() _default_resolver.add(cls) def __init__(self, **properties: Any) -> None: ''' ''' super().__init__() self._property_values = {} self._unstable_default_values = {} self._unstable_themed_values = {} for name, value in properties.items(): # TODO: this would be better to handle in descriptors if value is Undefined or value is Intrinsic: continue setattr(self, name, value) initialized = set(properties.keys()) for name in self.properties(_with_props=True): # avoid set[] for deterministic behavior if name in initialized: continue desc = self.lookup(name) if desc.has_unstable_default(self): desc._get(self) # this fills-in `_unstable_*_values` self._initialized = True def __setattr__(self, name: str, value: Any) -> None: ''' Intercept attribute setting on HasProps in order to special case a few situations: * short circuit all property machinery for ``_private`` attributes * suggest similar attribute names on attribute errors Args: name (str) : the name of the attribute to set on this object value (obj) : the value to set Returns: None ''' if name.startswith("_"): return super().__setattr__(name, value) properties = self.properties(_with_props=True) if name in properties: return super().__setattr__(name, value) descriptor = getattr(self.__class__, name, None) if isinstance(descriptor, property): # Python property return super().__setattr__(name, value) self._raise_attribute_error_with_matches(name, properties) def __getattr__(self, name: str) -> Any: ''' Intercept attribute setting on HasProps in order to special case a few situations: * short circuit all property machinery for ``_private`` attributes * suggest similar attribute names on attribute errors Args: name (str) : the name of the attribute to set on this object Returns: Any ''' if name.startswith("_"): return super().__getattribute__(name) properties = self.properties(_with_props=True) if name in properties: return super().__getattribute__(name) descriptor = getattr(self.__class__, name, None) if isinstance(descriptor, property): # Python property return super().__getattribute__(name) self._raise_attribute_error_with_matches(name, properties) def _raise_attribute_error_with_matches(self, name: str, properties: Iterable[str]) -> NoReturn: matches, text = difflib.get_close_matches(name.lower(), properties), "similar" if not matches: matches, text = sorted(properties), "possible" raise AttributeError(f"unexpected attribute {name!r} to {self.__class__.__name__}, {text} attributes are {nice_join(matches)}") def __str__(self) -> str: name = self.__class__.__name__ return f"{name}(...)" __repr__ = __str__ # Unfortunately we cannot implement __eq__. We rely on the default __hash__ # based on object identity, in order to put HasProps instances in sets. # Implementing __eq__ as structural equality would necessitate a __hash__ # that returns the same value different HasProps instances that compare # equal [1], and this would break many things. # # [1] https://docs.python.org/3/reference/datamodel.html#object.__hash__ # def equals(self, other: HasProps) -> bool: ''' Structural equality of models. Args: other (HasProps) : the other instance to compare to Returns: True, if properties are structurally equal, otherwise False ''' if not isinstance(other, self.__class__): return False else: return self.properties_with_values() == other.properties_with_values() def to_serializable(self, serializer: Serializer) -> ObjectRep: rep = ObjectRep( type="object", name=self.__qualified_model__, ) properties = self.properties_with_values(include_defaults=False) attributes = {key: serializer.encode(val) for key, val in properties.items()} if attributes: rep["attributes"] = attributes return rep # FQ type name required to suppress Sphinx error "more than one target found for cross-reference 'JSON'" def set_from_json(self, name: str, value: Any, *, setter: Setter | None = None) -> None: ''' Set a property value on this object from JSON. Args: name: (str) : name of the attribute to set json: (JSON-value) : value to set to the attribute to models (dict or None, optional) : Mapping of model ids to models (default: None) This is needed in cases where the attributes to update also have values that have references. setter(ClientSession or ServerSession or None, optional) : This is used to prevent "boomerang" updates to Bokeh apps. In the context of a Bokeh server application, incoming updates to properties will be annotated with the session that is doing the updating. This value is propagated through any subsequent change notifications that the update triggers. The session can compare the event setter to itself, and suppress any updates that originate from itself. Returns: None ''' if name in self.properties(_with_props=True): log.trace(f"Patching attribute {name!r} of {self!r} with {value!r}") # type: ignore # TODO: log.trace() descriptor = self.lookup(name) descriptor.set_from_json(self, value, setter=setter) else: log.warning("JSON had attr %r on obj %r, which is a client-only or invalid attribute that shouldn't have been sent", name, self) def update(self, **kwargs: Any) -> None: ''' Updates the object's properties from the given keyword arguments. Returns: None Examples: The following are equivalent: .. code-block:: python from bokeh.models import Range1d r = Range1d # set properties individually: r.start = 10 r.end = 20 # update properties together: r.update(start=10, end=20) ''' for k, v in kwargs.items(): setattr(self, k, v) @overload @classmethod def lookup(cls, name: str, *, raises: Literal[True] = True) -> PropertyDescriptor[Any]: ... @overload @classmethod def lookup(cls, name: str, *, raises: Literal[False] = False) -> PropertyDescriptor[Any] | None: ... @classmethod def lookup(cls, name: str, *, raises: bool = True) -> PropertyDescriptor[Any] | None: ''' Find the ``PropertyDescriptor`` for a Bokeh property on a class, given the property name. Args: name (str) : name of the property to search for raises (bool) : whether to raise or return None if missing Returns: PropertyDescriptor : descriptor for property named ``name`` ''' attr = getattr(cls, name, None) if attr is not None or (attr is None and not raises): return attr raise AttributeError(f"{cls.__name__}.{name} property descriptor does not exist") @overload @classmethod @lru_cache(None) def properties(cls, *, _with_props: Literal[False] = False) -> set[str]: ... @overload @classmethod @lru_cache(None) def properties(cls, *, _with_props: Literal[True] = True) -> dict[str, Property[Any]]: ... @classmethod @lru_cache(None) def properties(cls, *, _with_props: bool = False) -> set[str] | dict[str, Property[Any]]: ''' Collect the names of properties on this class. .. warning:: In a future version of Bokeh, this method will return a dictionary mapping property names to property objects. To future-proof this current usage of this method, wrap the return value in ``list``. Returns: property names ''' props: dict[str, Property[Any]] = {} for c in reversed(cls.__mro__): props.update(getattr(c, "__properties__", {})) if not _with_props: return set(props) return props @classmethod @lru_cache(None) def descriptors(cls) -> list[PropertyDescriptor[Any]]: """ List of property descriptors in the order of definition. """ return [ cls.lookup(name) for name, _ in cls.properties(_with_props=True).items() ] @classmethod @lru_cache(None) def properties_with_refs(cls) -> dict[str, Property[Any]]: ''' Collect the names of all properties on this class that also have references. This method *always* traverses the class hierarchy and includes properties defined on any parent classes. Returns: set[str] : names of properties that have references ''' return {k: v for k, v in cls.properties(_with_props=True).items() if v.has_ref} @classmethod @lru_cache(None) def dataspecs(cls) -> dict[str, DataSpec]: ''' Collect the names of all ``DataSpec`` properties on this class. This method *always* traverses the class hierarchy and includes properties defined on any parent classes. Returns: set[str] : names of ``DataSpec`` properties ''' from .property.dataspec import DataSpec # avoid circular import return {k: v for k, v in cls.properties(_with_props=True).items() if isinstance(v, DataSpec)} def properties_with_values(self, *, include_defaults: bool = True, include_undefined: bool = False) -> dict[str, Any]: ''' Collect a dict mapping property names to their values. This method *always* traverses the class hierarchy and includes properties defined on any parent classes. Non-serializable properties are skipped and property values are in "serialized" format which may be slightly different from the values you would normally read from the properties; the intent of this method is to return the information needed to losslessly reconstitute the object instance. Args: include_defaults (bool, optional) : Whether to include properties that haven't been explicitly set since the object was created. (default: True) Returns: dict : mapping from property names to their values ''' return self.query_properties_with_values(lambda prop: prop.serialized, include_defaults=include_defaults, include_undefined=include_undefined) @classmethod def _overridden_defaults(cls) -> dict[str, Any]: ''' Returns a dictionary of defaults that have been overridden. .. note:: This is an implementation detail of ``Property``. ''' defaults: dict[str, Any] = {} for c in reversed(cls.__mro__): defaults.update(getattr(c, "__overridden_defaults__", {})) return defaults def query_properties_with_values(self, query: Callable[[PropertyDescriptor[Any]], bool], *, include_defaults: bool = True, include_undefined: bool = False) -> dict[str, Any]: ''' Query the properties values of |HasProps| instances with a predicate. Args: query (callable) : A callable that accepts property descriptors and returns True or False include_defaults (bool, optional) : Whether to include properties that have not been explicitly set by a user (default: True) Returns: dict : mapping of property names and values for matching properties ''' themed_keys: set[str] = set() result: dict[str, Any] = {} keys = self.properties(_with_props=True) if include_defaults: selected_keys = set(keys) else: # TODO (bev) For now, include unstable default values. Things rely on Instances # always getting serialized, even defaults, and adding unstable defaults here # accomplishes that. Unmodified defaults for property value containers will be # weeded out below. selected_keys = set(self._property_values.keys()) | set(self._unstable_default_values.keys()) themed_values = self.themed_values() if themed_values is not None: themed_keys = set(themed_values.keys()) selected_keys |= themed_keys for key in keys: descriptor = self.lookup(key) if not query(descriptor): continue try: value = descriptor.get_value(self) except UnsetValueError: if include_undefined: value = Undefined else: raise else: # TODO: this should happen before get_value(), however there's currently # no reliable way of checking if a property is unset without actually # getting the value. if key not in selected_keys: continue if not include_defaults and key not in themed_keys: if isinstance(value, PropertyValueContainer) and key in self._unstable_default_values: continue result[key] = value return result def themed_values(self) -> dict[str, Any] | None: ''' Get any theme-provided overrides. Results are returned as a dict from property name to value, or ``None`` if no theme overrides any values for this instance. Returns: dict or None ''' return getattr(self, '__themed_values__', None) def apply_theme(self, property_values: dict[str, Any]) -> None: ''' Apply a set of theme values which will be used rather than defaults, but will not override application-set values. The passed-in dictionary may be kept around as-is and shared with other instances to save memory (so neither the caller nor the |HasProps| instance should modify it). Args: property_values (dict) : theme values to use in place of defaults Returns: None ''' old_dict = self.themed_values() # if the same theme is set again, it should reuse the same dict if old_dict is property_values: # lgtm [py/comparison-using-is] return removed: set[str] = set() # we're doing a little song-and-dance to avoid storing __themed_values__ or # an empty dict, if there's no theme that applies to this HasProps instance. if old_dict is not None: removed.update(set(old_dict.keys())) added = set(property_values.keys()) old_values: dict[str, Any] = {} for k in added.union(removed): old_values[k] = getattr(self, k) if len(property_values) > 0: setattr(self, '__themed_values__', property_values) elif hasattr(self, '__themed_values__'): delattr(self, '__themed_values__') # Property container values might be cached even if unmodified. Invalidate # any cached values that are not modified at this point. for k, v in old_values.items(): if k in self._unstable_themed_values: del self._unstable_themed_values[k] # Emit any change notifications that result for k, v in old_values.items(): descriptor = self.lookup(k) if isinstance(descriptor, PropertyDescriptor): descriptor.trigger_if_changed(self, v) def unapply_theme(self) -> None: ''' Remove any themed values and restore defaults. Returns: None ''' self.apply_theme(property_values={}) def clone(self) -> HasProps: ''' Duplicate a HasProps object. This creates a shallow clone of the original model, i.e. any mutable containers or child models will not be duplicated. ''' attrs = self.properties_with_values(include_defaults=False, include_undefined=True) return self.__class__(**{key: val for key, val in attrs.items() if val is not Undefined}) KindRef = Any # TODO class PropertyDef(TypedDict): name: str kind: KindRef default: NotRequired[Any] class OverrideDef(TypedDict): name: str default: Any class ModelDef(TypedDict): type: Literal["model"] name: str extends: NotRequired[Ref | None] properties: NotRequired[list[PropertyDef]] overrides: NotRequired[list[OverrideDef]] def _HasProps_to_serializable(cls: type[HasProps], serializer: Serializer) -> Ref | ModelDef: from ..model import DataModel, Model ref = Ref(id=ID(cls.__qualified_model__)) serializer.add_ref(cls, ref) if not is_DataModel(cls): return ref # TODO: consider supporting mixin models bases: list[type[HasProps]] = [ base for base in cls.__bases__ if issubclass(base, Model) and base != DataModel ] if len(bases) == 0: extends = None elif len(bases) == 1: [base] = bases extends = serializer.encode(base) else: serializer.error("multiple bases are not supported") properties: list[PropertyDef] = [] overrides: list[OverrideDef] = [] # TODO: don't use unordered sets for prop_name in cls.__properties__: descriptor = cls.lookup(prop_name) kind = "Any" # TODO: serialize kinds default = descriptor.property._default if default is Undefined: prop_def = PropertyDef(name=prop_name, kind=kind) else: if descriptor.is_unstable(default): default = default() prop_def = PropertyDef(name=prop_name, kind=kind, default=serializer.encode(default)) properties.append(prop_def) for prop_name, default in getattr(cls, "__overridden_defaults__", {}).items(): overrides.append(OverrideDef(name=prop_name, default=serializer.encode(default))) modeldef = ModelDef( type="model", name=cls.__qualified_model__, ) if extends is not None: modeldef["extends"] = extends if properties: modeldef["properties"] = properties if overrides: modeldef["overrides"] = overrides return modeldef Serializer.register(MetaHasProps, _HasProps_to_serializable) #----------------------------------------------------------------------------- # Private API #----------------------------------------------------------------------------- _ABSTRACT_ADMONITION = ''' .. note:: This is an abstract base class used to help organize the hierarchy of Bokeh model types. **It is not useful to instantiate on its own.** ''' #----------------------------------------------------------------------------- # Code #-----------------------------------------------------------------------------