Added support for forward-references in dataclasses

runtype/dataclass.py

@@ -7,14 +7,21 @@
 import dataclasses
 from typing import Union
 from abc import ABC, abstractmethod
+import inspect
 
+from .utils import ForwardRef
 from .common import CHECK_TYPES
 from .validation import TypeMismatchError, ensure_isa as default_ensure_isa
-from .pytypes import cast_to_type, SumType, NoneType
+from .pytypes import TypeCaster, type_caster, SumType, NoneType
 
 Required = object()
 MAX_SAMPLE_SIZE = 16
 
+class NopTypeCaster:
+    cache = {}
+    def to_canon(self, t):
+        return t
+
 class Configuration(ABC):
     """Generic configuration template for dataclass. Mainly for type-checking.
 
@@ -48,15 +55,15 @@ class Form:
 
     """
 
-    def canonize_type(self, t):
-        """Given a type, return its canonical form.
+    def on_default(self, default):
+        """Called whenever a dataclass member is assigned a default value.
         """
-        return t
+        return default
 
-    def on_default(self, t, default):
-        """Called whenever a dataclass member is assigned a default value.
+    def make_type_caster(self, frame):
+        """Return a type caster, as defined in pytypes.TypeCaster
         """
-        return t, default
+        return NopTypeCaster()
 
     @abstractmethod
     def ensure_isa(self, a, b, sampler=None):
@@ -79,40 +86,48 @@ class PythonConfiguration(Configuration):
 
     This is the default class given to the ``dataclass()`` function.
     """
-    canonize_type = staticmethod(cast_to_type)
+    make_type_caster = TypeCaster
     ensure_isa = staticmethod(default_ensure_isa)
 
     def cast(self, obj, to_type):
         return to_type.cast_from(obj)
 
-    def on_default(self, type_, default):
-        if default is None:
-            type_ = SumType([type_, NoneType])
-        elif isinstance(default, (list, dict, set)):
+    def on_default(self, default):
+        if isinstance(default, (list, dict, set)):
             def f(_=default):
                 return copy(_)
-            default = dataclasses.field(default_factory=f)
-        return type_, default
-
+            return dataclasses.field(default_factory=f)
+        return default
 
 
-def _post_init(self, config, should_cast, sampler):
+def _post_init(self, config, should_cast, sampler, type_caster):
     for name, field in getattr(self, '__dataclass_fields__', {}).items():
         value = getattr(self, name)
 
         if value is Required:
             raise TypeError(f"Field {name} requires a value")
 
+        try:
+            type_ = type_caster.cache[id(field)]
+        except KeyError:
+            type_ = field.type
+            if isinstance(type_, str):
+                type_ = ForwardRef(type_)
+            type_ = type_caster.to_canon(type_)
+            if field.default is None:
+                type_ = SumType([type_, NoneType])
+            type_caster.cache[id(field)] = type_
+
         try:
             if should_cast:    # Basic cast
                 assert not sampler
-                value = config.cast(value, field.type)
+                value = config.cast(value, type_)
                 object.__setattr__(self, name, value)
             else:
-                config.ensure_isa(value, field.type, sampler)
+                config.ensure_isa(value, type_, sampler)
         except TypeMismatchError as e:
             item_value, item_type = e.args
-            msg = f"[{type(self).__name__}] Attribute '{name}' expected value of type {field.type}."
+            msg = f"[{type(self).__name__}] Attribute '{name}' expected value of type '{type_}'."
             msg += f" Instead got {value!r}"
             if item_value is not value:
                 msg += f'\n\n    Failed on item: {item_value!r}, expected type {item_type}'
@@ -197,9 +212,9 @@ def _sample(seq, max_sample_size=MAX_SAMPLE_SIZE):
         return seq
     return random.sample(seq, max_sample_size)
 
-def _process_class(cls, config, check_types, **kw):
+def _process_class(cls, config, check_types, context_frame, **kw):
     for name, type_ in getattr(cls, '__annotations__', {}).items():
-        type_ = config.canonize_type(type_)
+        # type_ = config.type_to_canon(type_) if not isinstance(type_, str) else type_
 
         # If default not specified, assign Required, for a later check
         # We don't assign MISSING; we want to bypass dataclass which is too strict for this
@@ -211,7 +226,7 @@ def _process_class(cls, config, check_types, **kw):
                 if default.default is dataclasses.MISSING and default.default_factory is dataclasses.MISSING:
                     default.default = Required
 
-            type_, new_default = config.on_default(type_, default)
+            new_default = config.on_default(default)
             if new_default is not default:
                 setattr(cls, name, new_default)
 
@@ -222,9 +237,12 @@ def _process_class(cls, config, check_types, **kw):
 
         orig_post_init = getattr(cls, '__post_init__', None)
         sampler = _sample if check_types=='sample' else None
+        # eval_type_string = EvalInContext(context_frame)
+        type_caster = config.make_type_caster(context_frame)
 
         def __post_init__(self):
-            _post_init(self, config=config, should_cast=check_types == 'cast', sampler=sampler)
+            # Only now context_frame has complete information
+            _post_init(self, config=config, should_cast=check_types == 'cast', sampler=sampler, type_caster=type_caster)
             if orig_post_init is not None:
                 orig_post_init(self)
 
@@ -340,8 +358,9 @@ def dataclass(cls=None, *, check_types: Union[bool, str] = CHECK_TYPES,
     """
     assert isinstance(config, Configuration)
 
+    context_frame = inspect.currentframe().f_back   # Get parent frame, to resolve forward-references
     def wrap(cls):
-        return _process_class(cls, config, check_types,
+        return _process_class(cls, config, check_types, context_frame,
                               init=init, repr=repr, eq=eq, order=order, unsafe_hash=unsafe_hash, frozen=frozen, slots=slots)
 
     # See if we're being called as @dataclass or @dataclass().

runtype/pytypes.py

@@ -9,6 +9,7 @@
 import typing
 from datetime import datetime
 
+from .utils import ForwardRef
 from .base_types import DataType, Validator, TypeMismatchError
 from . import base_types
 from . import datetime_parse
@@ -27,7 +28,7 @@ class PythonType(base_types.Type, Validator):
 
 class Constraint(base_types.Constraint):
     def __init__(self, for_type, predicates):
-        super().__init__(cast_to_type(for_type), predicates)
+        super().__init__(type_caster.to_canon(for_type), predicates)
 
     def cast_from(self, obj):
         obj = self.type.cast_from(obj)
@@ -194,7 +195,7 @@ def __init__(self, base, item=Any*Any):
         super().__init__(base)
         if isinstance(item, tuple):
             assert len(item) == 2
-            item = ProductType([cast_to_type(x) for x in item])
+            item = ProductType([type_caster.to_canon(x) for x in item])
         self.item = item
 
     def validate_instance(self, obj, sampler=None):
@@ -329,85 +330,99 @@ def cast_from(self, obj):
     origin_frozenset = typing.FrozenSet
 
 
-def _cast_to_type(t):
-    if isinstance(t, Validator):
-        return t
+class TypeCaster:
+    def __init__(self, frame=None):
+        self.cache = {}
+        self.frame = frame
 
-    if isinstance(t, tuple):
-        return SumType([cast_to_type(x) for x in t])
+    def _to_canon(self, t):
+        to_canon = self.to_canon
+
+        if isinstance(t, (base_types.Type, Validator)):
+            return t
+
+        if isinstance(t, ForwardRef):
+            t = t._evaluate(self.frame.f_globals, self.frame.f_locals, set())
+
+        if isinstance(t, tuple):
+            return SumType([to_canon(x) for x in t])
+
+        try:
+            t.__origin__
+        except AttributeError:
+            pass
+        else:
+            if getattr(t, '__args__', None) is None:
+                if t is typing.List:
+                    return List
+                elif t is typing.Dict:
+                    return Dict
+                elif t is typing.Set:
+                    return Set
+                elif t is typing.FrozenSet:
+                    return FrozenSet
+                elif t is typing.Tuple:
+                    return Tuple
+                elif t is typing.Mapping:  # 3.6
+                    return Mapping
+                elif t is typing.Sequence:
+                    return Sequence
+
+            if t.__origin__ is origin_list:
+                x ,= t.__args__
+                return List[to_canon(x)]
+            elif t.__origin__ is origin_set:
+                x ,= t.__args__
+                return Set[to_canon(x)]
+            elif t.__origin__ is origin_frozenset:
+                x ,= t.__args__
+                return FrozenSet[to_canon(x)]
+            elif t.__origin__ is origin_dict:
+                k, v = t.__args__
+                return Dict[to_canon(k), to_canon(v)]
+            elif t.__origin__ is origin_tuple:
+                if Ellipsis in t.__args__:
+                    if len(t.__args__) != 2 or t.__args__[0] == Ellipsis:
+                        raise ValueError("Tuple with '...'' expected to be of the exact form: tuple[t, ...].")
+                    return TupleEllipsis[to_canon(t.__args__[0])]
+
+                return ProductType([to_canon(x) for x in t.__args__])
+
+            elif t.__origin__ is typing.Union:
+                return SumType([to_canon(x) for x in t.__args__])
+            elif t.__origin__ is abc.Callable or t is typing.Callable:
+                # return Callable[ProductType(to_canon(x) for x in t.__args__)]
+                return Callable  # TODO
+            elif py38 and t.__origin__ is typing.Literal:
+                return OneOf(t.__args__)
+            elif t.__origin__ is abc.Mapping or t.__origin__ is typing.Mapping:
+                k, v = t.__args__
+                return Mapping[to_canon(k), to_canon(v)]
+            elif t.__origin__ is abc.Sequence or t.__origin__ is typing.Sequence:
+                x ,= t.__args__
+                return Sequence[to_canon(x)]
+
+            elif t.__origin__ is type or t.__origin__ is typing.Type:
+                # TODO test issubclass on t.__args__
+                return PythonDataType(type)
+
+            raise NotImplementedError("No support for type:", t)
+
+        if isinstance(t, typing.TypeVar):
+            return Any  # XXX is this correct?
+
+        return PythonDataType(t)
+
+    def to_canon(self, t):
+        try:
+            return self.cache[t]
+        except KeyError:
+            try:
+                res = _type_cast_mapping[t]
+            except KeyError:
+                res = self._to_canon(t)
+            self.cache[t] = res     # memoize
+            return res
 
-    try:
-        t.__origin__
-    except AttributeError:
-        pass
-    else:
-        if getattr(t, '__args__', None) is None:
-            if t is typing.List:
-                return List
-            elif t is typing.Dict:
-                return Dict
-            elif t is typing.Set:
-                return Set
-            elif t is typing.FrozenSet:
-                return FrozenSet
-            elif t is typing.Tuple:
-                return Tuple
-            elif t is typing.Mapping:  # 3.6
-                return Mapping
-            elif t is typing.Sequence:
-                return Sequence
-
-        if t.__origin__ is origin_list:
-            x ,= t.__args__
-            return List[cast_to_type(x)]
-        elif t.__origin__ is origin_set:
-            x ,= t.__args__
-            return Set[cast_to_type(x)]
-        elif t.__origin__ is origin_frozenset:
-            x ,= t.__args__
-            return FrozenSet[cast_to_type(x)]
-        elif t.__origin__ is origin_dict:
-            k, v = t.__args__
-            return Dict[cast_to_type(k), cast_to_type(v)]
-        elif t.__origin__ is origin_tuple:
-            if Ellipsis in t.__args__:
-                if len(t.__args__) != 2 or t.__args__[0] == Ellipsis:
-                    raise ValueError("Tuple with '...'' expected to be of the exact form: tuple[t, ...].")
-                return TupleEllipsis[cast_to_type(t.__args__[0])]
-
-            return ProductType([cast_to_type(x) for x in t.__args__])
-
-        elif t.__origin__ is typing.Union:
-            return SumType([cast_to_type(x) for x in t.__args__])
-        elif t.__origin__ is abc.Callable or t is typing.Callable:
-            # return Callable[ProductType(cast_to_type(x) for x in t.__args__)]
-            return Callable  # TODO
-        elif py38 and t.__origin__ is typing.Literal:
-            return OneOf(t.__args__)
-        elif t.__origin__ is abc.Mapping or t.__origin__ is typing.Mapping:
-            k, v = t.__args__
-            return Mapping[cast_to_type(k), cast_to_type(v)]
-        elif t.__origin__ is abc.Sequence or t.__origin__ is typing.Sequence:
-            x ,= t.__args__
-            return Sequence[_cast_to_type(x)]
-
-        elif t.__origin__ is type or t.__origin__ is typing.Type:
-            # TODO test issubclass on t.__args__
-            return PythonDataType(type)
-
-        raise NotImplementedError("No support for type:", t)
-
-    if isinstance(t, typing.TypeVar):
-        return Any  # XXX is this correct?
-
-    return PythonDataType(t)
-
-
-def cast_to_type(t):
-    try:
-        return _type_cast_mapping[t]
-    except KeyError:
-        res = _cast_to_type(t)
-        _type_cast_mapping[t] = res     # memoize
-        return res
 
+type_caster = TypeCaster()

runtype/utils.py

@@ -1,4 +1,22 @@
 import inspect
+import sys
+
+if sys.version_info < (3, 7):
+    # python 3.6 
+    from typing import _ForwardRef as ForwardRef
+    _orig_eval = ForwardRef._eval_type
+elif sys.version_info < (3, 9):
+    from typing import ForwardRef
+    _orig_eval = ForwardRef._evaluate
+else:
+    from typing import ForwardRef
+
+if sys.version_info < (3, 9):
+    def _evaluate(self, g, l, _):
+        return _orig_eval(self, g, l)
+    ForwardRef._evaluate = _evaluate
+
+
 
 def get_func_signatures(typesystem, f):
     sig = inspect.signature(f)