def test_simple_value_word(self, simple_value_word) -> None:
_, effect = concat.typecheck.infer(
concat.typecheck.Environment(),
[simple_value_word],
initial_stack=TypeSequence([]),
)
expected_types = {
concat.parse.NoneWordNode: none_type,
concat.parse.NotImplWordNode: not_implemented_type,
concat.parse.EllipsisWordNode: ellipsis_type,
}
expected_type = expected_types[type(simple_value_word)]
self.assertEqual(list(effect.output), [expected_type])
def to_type(self, env: Environment) -> Tuple[TypeSequence, Environment]:
sequence: List[StackItemType] = []
if self._sequence_variable is None:
# implicit stack polymorphism
sequence.append(SequenceVariable())
elif self._sequence_variable not in env:
env = env.copy()
env[self._sequence_variable] = SequenceVariable()
sequence.append(env[self._sequence_variable])
for type_node in self._individual_type_items:
type, env = type_node.to_type(env)
sequence.append(type)
return TypeSequence(sequence), env
def test_attribute_word(self, attr_word) -> None:
_, type = concat.typecheck.infer(
concat.typecheck.Environment(),
[attr_word],
initial_stack=TypeSequence(
[
ObjectType(
IndividualVariable(),
{attr_word.value: StackEffect([], [int_type]),},
),
]
),
)
self.assertEqual(list(type.output), [int_type])
def test_with_word_and_funcdef_inference(self) -> None:
wth = 'def fn(f:object -- n:int): drop 0 ~\n$fn {"file": "a_file"} open with'
tree = parse(wth)
in_var = SequenceVariable()
_, type = concat.typecheck.infer(
Environment(
{
**concat.typecheck.preamble_types.types,
'drop': concat.typecheck.types.ForAll(
[in_var], StackEffect([in_var, object_type], [in_var])
),
'open': concat.typecheck.types.ForAll(
[in_var],
# FIXME: dict_type should be a type constructor
StackEffect([in_var, dict_type], [in_var, file_type]),
),
}
),
tree.children,
initial_stack=TypeSequence([in_var]),
)
self.assertEqual(type, StackEffect([in_var], [in_var, int_type]))
def test_class_subtype_of_py_function(self, type1, type2) -> None:
x = IndividualVariable()
py_function = py_function_type[TypeSequence([type1]), type2]
unbound_py_function = py_function_type[TypeSequence([x, type1]), type2]
cls = ClassType(x, {'__init__': unbound_py_function})
self.assertLessEqual(cls, py_function)
def test_object_subtype_of_py_function(self, type1, type2) -> None:
x = IndividualVariable()
py_function = py_function_type[TypeSequence([type1]), type2]
object = ObjectType(x, {'__call__': py_function})
self.assertLessEqual(object, py_function)
_stack_var = SequenceVariable()
_stack_type_var = SequenceVariable()
_a_var = IndividualVariable()
_b_var = IndividualVariable()
_c_var = IndividualVariable()
_x = IndividualVariable()
types = {
'py_call': ForAll(
[_rest_var, _seq_var, _a_var],
StackEffect(
[
_rest_var,
iterable_type[object_type,],
iterable_type[object_type,],
py_function_type[TypeSequence([_seq_var]), _a_var],
],
[_rest_var, _a_var],
),
),
'swap': ForAll(
[_rest_var, _a_var, _b_var],
StackEffect([_rest_var, _a_var, _b_var], [_rest_var, _b_var, _a_var]),
),
'pick': ForAll(
[_rest_var, _a_var, _b_var, _c_var],
StackEffect(
[_rest_var, _a_var, _b_var, _c_var],
[_rest_var, _a_var, _b_var, _c_var, _a_var],
),
),
def infer(
gamma: Environment,
e: 'concat.astutils.WordsOrStatements',
extensions: Optional[Tuple[Callable]] = None,
is_top_level=False,
source_dir='.',
initial_stack: Optional[TypeSequence] = None,
) -> Tuple[Substitutions, StackEffect]:
"""The infer function described by Kleffner."""
e = list(e)
current_subs = Substitutions()
if initial_stack is None:
initial_stack = TypeSequence(
[] if is_top_level else [SequenceVariable()])
current_effect = StackEffect(initial_stack, initial_stack)
for node in e:
try:
S, (i, o) = current_subs, current_effect
if isinstance(node, concat.operators.AddWordNode):
# rules:
# require object_type because the methods should return
# NotImplemented for most types
# FIXME: Make the rules safer... somehow
# ... a b => (... {__add__(object) -> s} t)
# ---
# a b + => (... s)
# ... a b => (... t {__radd__(object) -> s})
# ---
# a b + => (... s)
*rest, type1, type2 = current_effect.output
try_radd = False
try:
add_type = type1.get_type_of_attribute('__add__')
except AttributeError:
try_radd = True
else:
if not isinstance(add_type, ObjectType):
raise TypeError(
'__add__ method of type {} is not of an object type, instead has type {}'
.format(type1, add_type))
if add_type.head != py_function_type:
raise TypeError(
'__add__ method of type {} is not a Python function, instead it has type {}'
.format(type1, add_type))
if [*add_type.type_arguments[0]] != [object_type]:
raise TypeError(
'__add__ method of type {} does not have type (object) -> `t, instead it has type {}'
.format(type1, add_type))
current_effect = StackEffect(
current_effect.input,
[*rest, add_type.output],
)
if try_radd:
radd_type = type2.get_type_of_attribute('__radd__')
if (not isinstance(radd_type, ObjectType)
or radd_type.head != py_function_type or
[*radd_type.type_arguments[0]] != [object_type]):
raise TypeError(
'__radd__ method of type {} does not have type (object) -> `t, instead it has type {} (left operand is of type {})'
.format(type2, radd_type, type1))
current_effect = StackEffect(
current_effect.input,
[*rest, radd_type.output],
)
elif isinstance(node, concat.parse.PushWordNode):
S1, (i1, o1) = S, (i, o)
# special case for pushing an attribute accessor
child = node.children[0]
if isinstance(child, concat.parse.AttributeWordNode):
top = o1[-1]
attr_type = top.get_type_of_attribute(child.value)
rest_types = o1[:-1]
current_subs, current_effect = (
S1,
StackEffect(i1, [*rest_types, attr_type]),
)
# special case for name words
elif isinstance(child, concat.parse.NameWordNode):
if child.value not in gamma:
raise NameError(child)
name_type = gamma[child.value].instantiate()
current_effect = StackEffect(
current_effect.input,
[*current_effect.output,
current_subs(name_type)],
)
elif isinstance(child, concat.parse.SliceWordNode):
sliceable_object_type = o[-1]
# This doesn't match the evaluation order used by the
# transpiler.
# FIXME: Change the transpiler to fit the type checker.
sub1, start_effect = infer(
gamma,
list(child.start_children),
extensions=extensions,
source_dir=source_dir,
initial_stack=TypeSequence(o[:-1]),
)
start_type = start_effect.output[-1]
o = tuple(start_effect.output[:-1])
sub2, stop_effect = infer(
sub1(gamma),
list(child.stop_children),
extensions=extensions,
source_dir=source_dir,
initial_stack=TypeSequence(o),
)
stop_type = stop_effect.output[-1]
o = tuple(stop_effect.output[:-1])
sub3, step_effect = infer(
sub2(sub1(gamma)),
list(child.step_children),
extensions=extensions,
source_dir=source_dir,
initial_stack=TypeSequence(o),
)
step_type = step_effect.output[-1]
o = tuple(step_effect.output[:-1])
this_slice_type = slice_type[start_type, stop_type,
step_type]
getitem_type = sliceable_object_type.get_type_of_attribute(
'__getitem__')
getitem_type = getitem_type.get_type_of_attribute(
'__call__')
getitem_type = getitem_type.instantiate()
if (not isinstance(getitem_type, PythonFunctionType)
or len(getitem_type.input) != 1):
raise TypeError(
'__getitem__ method of {} has incorrect type {}'.
format(node, getitem_type))
getitem_type, overload_subs = getitem_type.select_overload(
(this_slice_type, ))
result_type = getitem_type.output
current_subs = overload_subs(sub3(sub2(
sub1(current_subs))))
current_effect = current_subs(
StackEffect(i, [*o, result_type]))
# special case for subscription words
elif isinstance(child, concat.parse.SubscriptionWordNode):
S2, (i2, o2) = infer(
current_subs(gamma),
child.children,
extensions=extensions,
is_top_level=False,
source_dir=source_dir,
initial_stack=current_effect.output,
)
# FIXME: Should be generic
subscriptable_interface = subscriptable_type[
int_type, IndividualVariable(), ]
rest_var = SequenceVariable()
expected_o2 = TypeSequence([
rest_var,
subscriptable_interface,
int_type,
])
o2[-1].constrain(int_type)
getitem_type = (o2[-2].get_type_of_attribute(
'__getitem__').instantiate().get_type_of_attribute(
'__call__').instantiate())
if not isinstance(getitem_type, PythonFunctionType):
raise TypeError(
'__getitem__ of type {} is not a Python function (has type {})'
.format(o2[-2], getitem_type))
getitem_type, overload_subs = getitem_type.select_overload(
[int_type])
current_subs = overload_subs(S2(current_subs))
current_effect = current_subs(
StackEffect(
current_effect.input,
[*o2[:-2], getitem_type.output],
))
else:
if (isinstance(child, concat.parse.QuoteWordNode)
and child.input_stack_type is not None):
input_stack, _ = child.input_stack_type.to_type(gamma)
else:
# The majority of quotations I've written don't comsume
# anything on the stack, so make that the default.
input_stack = TypeSequence([SequenceVariable()])
S2, fun_type = infer(
S1(gamma),
child.children,
extensions=extensions,
source_dir=source_dir,
initial_stack=input_stack,
)
current_subs, current_effect = (
S2(S1),
StackEffect(
S2(TypeSequence(i1)),
[*S2(TypeSequence(o1)),
QuotationType(fun_type)],
),
)
elif isinstance(node, concat.parse.WithWordNode):
a_bar, b_bar = SequenceVariable(), SequenceVariable()
body_type = StackEffect([a_bar, object_type], [b_bar])
phi = current_effect.output.constrain_and_bind_supertype_variables(
TypeSequence([a_bar, body_type, context_manager_type]),
set(),
)
assert b_bar in phi
current_subs, current_effect = (
phi(current_subs),
phi(
StackEffect(current_effect.input,
TypeSequence([b_bar]))),
)
elif isinstance(node, concat.parse.TryWordNode):
a_bar, b_bar = SequenceVariable(), SequenceVariable()
phi = TypeSequence(o).constrain_and_bind_supertype_variables(
TypeSequence([
a_bar,
iterable_type[StackEffect([a_bar], [b_bar]), ],
StackEffect([a_bar], [b_bar]),
]),
set(),
)
assert b_bar in phi
current_subs, current_effect = (
phi(S),
phi(StackEffect(i, [b_bar])),
)
elif isinstance(node, concat.parse.DictWordNode):
phi = current_subs
collected_type = current_effect.output
for key, value in node.dict_children:
phi1, (i1, o1) = infer(
phi(gamma),
key,
extensions=extensions,
source_dir=source_dir,
initial_stack=collected_type,
)
phi = phi1(phi)
collected_type = phi(o1)
# drop the top of the stack to use as the key
collected_type, key_type = (
collected_type[:-1],
collected_type.as_sequence()[-1],
)
phi2, (i2, o2) = infer(
phi(gamma),
value,
extensions=extensions,
source_dir=source_dir,
initial_stack=collected_type,
)
phi = phi2(phi)
collected_type = phi(o2)
# drop the top of the stack to use as the value
collected_type, value_type = (
collected_type[:-1],
collected_type.as_sequence()[-1],
)
current_subs, current_effect = (
phi,
phi(
StackEffect(current_effect.input,
[*collected_type, dict_type])),
)
elif isinstance(node, concat.parse.ListWordNode):
phi = S
collected_type = TypeSequence(o)
element_type: IndividualType = object_type
for item in node.list_children:
phi1, fun_type = infer(
phi(gamma),
item,
extensions=extensions,
source_dir=source_dir,
initial_stack=collected_type,
)
collected_type = fun_type.output
# FIXME: Infer the type of elements in the list based on
# ALL the elements.
if element_type == object_type:
assert isinstance(collected_type[-1], IndividualType)
element_type = collected_type[-1]
# drop the top of the stack to use as the item
collected_type = collected_type[:-1]
phi = phi1(phi)
current_subs, current_effect = (
phi,
phi(
StackEffect(
i, [*collected_type, list_type[element_type, ]])),
)
elif isinstance(node, concat.operators.InvertWordNode):
out_types = current_effect.output[:-1]
invert_attr_type = current_effect.output[
-1].get_type_of_attribute('__invert__')
if not isinstance(invert_attr_type, PythonFunctionType):
raise TypeError(
'__invert__ of type {} must be a Python function'.
format(current_effect.output[-1]))
result_type = invert_attr_type.output
current_effect = StackEffect(current_effect.input,
[*out_types, result_type])
elif isinstance(node, concat.parse.NoneWordNode):
current_effect = StackEffect(i, [*o, none_type])
elif isinstance(node, concat.parse.NotImplWordNode):
current_effect = StackEffect(i, [*o, not_implemented_type])
elif isinstance(node, concat.parse.EllipsisWordNode):
current_effect = StackEffect(i, [*o, ellipsis_type])
elif isinstance(node, concat.parse.SliceWordNode):
sliceable_object_type = o[-1]
# This doesn't match the evaluation order used by the
# transpiler.
# FIXME: Change the transpiler to fit the type checker.
sub1, start_effect = infer(
gamma,
list(node.start_children),
source_dir=source_dir,
initial_stack=TypeSequence(o[:-1]),
)
start_type = start_effect.output[-1]
o = tuple(start_effect.output[:-1])
sub2, stop_effect = infer(
sub1(gamma),
list(node.stop_children),
source_dir=source_dir,
initial_stack=TypeSequence(o),
)
stop_type = stop_effect.output[-1]
o = tuple(stop_effect.output[:-1])
sub3, step_effect = infer(
sub2(sub1(gamma)),
list(node.step_children),
source_dir=source_dir,
initial_stack=TypeSequence(o),
)
step_type = step_effect.output[-1]
o = tuple(step_effect.output[:-1])
this_slice_type = slice_type[start_type, stop_type, step_type]
getitem_type = sliceable_object_type.get_type_of_attribute(
'__getitem__')
getitem_type = getitem_type.get_type_of_attribute('__call__')
getitem_type = getitem_type.instantiate()
if (not isinstance(getitem_type, PythonFunctionType)
or len(getitem_type.input) != 1):
raise TypeError(
'__getitem__ method of {} has incorrect type {}'.
format(node, getitem_type))
getitem_type, overload_subs = getitem_type.select_overload(
(this_slice_type, ))
result_type = getitem_type.output
current_subs = overload_subs(sub3(sub2(sub1(current_subs))))
current_effect = overload_subs(
StackEffect(i, [*o, result_type]))
elif isinstance(node, concat.parse.FromImportStatementNode):
imported_name = node.asname or node.imported_name
# mutate type environment
gamma[imported_name] = object_type
# We will try to find a more specific type.
sys.path, old_path = [source_dir, *sys.path], sys.path
module = importlib.import_module(node.value)
sys.path = old_path
# For now, assume the module's written in Python.
try:
gamma[imported_name] = current_subs(
getattr(module, '@@types')[node.imported_name])
except (KeyError, builtins.AttributeError):
# attempt introspection to get a more specific type
if callable(getattr(module, node.imported_name)):
args_var = SequenceVariable()
gamma[imported_name] = ObjectType(
IndividualVariable(),
{
'__call__':
py_function_type[TypeSequence([args_var]),
object_type],
},
type_parameters=[args_var],
nominal=True,
)
elif isinstance(node, concat.parse.ImportStatementNode):
# TODO: Support all types of import correctly.
if node.asname is not None:
gamma[node.asname] = current_subs(
_generate_type_of_innermost_module(
node.value,
source_dir).generalized_wrt(current_subs(gamma)))
else:
imported_name = node.value
# mutate type environment
components = node.value.split('.')
# FIXME: This replaces whatever was previously imported. I really
# should implement namespaces properly.
gamma[components[0]] = current_subs(
_generate_module_type(components,
source_dir=source_dir))
elif isinstance(node, concat.parse.SubscriptionWordNode):
seq = current_effect.output[:-1]
index_type_var = IndividualVariable()
result_type_var = IndividualVariable()
subscriptable_interface = subscriptable_type[index_type_var,
result_type_var]
(
index_subs,
(index_input, index_output),
) = infer(
gamma,
node.children,
initial_stack=current_effect.output,
extensions=extensions,
)
index_output_typeseq = index_output
subs = index_output_typeseq[
-2].constrain_and_bind_supertype_variables(
subscriptable_interface,
set(),
)(index_subs(current_subs))
subs(index_output_typeseq[-1]).constrain(subs(index_type_var))
result_type = subs(result_type_var).get_type_of_attribute(
'__call__')
if not isinstance(result_type, StackEffect):
raise TypeError(
'result of subscription is not callable as a Concat function (has type {})'
.format(result_type))
subs = index_output_typeseq[:
-2].constrain_and_bind_supertype_variables(
result_type.input, set())(subs)
current_subs, current_effect = (
subs,
subs(StackEffect(current_effect.input,
result_type.output)),
)
elif isinstance(node, concat.operators.SubtractWordNode):
# FIXME: We should check if the other operand supports __rsub__ if the
# first operand doesn't support __sub__.
other_operand_type_var = IndividualVariable()
result_type_var = IndividualVariable()
subtractable_interface = subtractable_type[
other_operand_type_var, result_type_var]
seq_var = SequenceVariable()
final_subs = current_effect.output.constrain_and_bind_supertype_variables(
TypeSequence([
seq_var,
subtractable_interface,
other_operand_type_var,
]),
set(),
)
assert seq_var in final_subs
current_subs, current_effect = (
final_subs(current_subs),
final_subs(
StackEffect(current_effect.input,
[seq_var, result_type_var])),
)
elif isinstance(node, concat.parse.FuncdefStatementNode):
S = current_subs
f = current_effect
name = node.name
declared_type: Optional[StackEffect]
if node.stack_effect:
declared_type, _ = node.stack_effect.to_type(S(gamma))
declared_type = S(declared_type)
else:
# NOTE: To continue the "bidirectional" bent, we will require a
# type annotation.
# TODO: Make the return types optional?
# FIXME: Should be a parse error.
raise TypeError(
'must have type annotation on function definition')
recursion_env = gamma.copy()
recursion_env[name] = declared_type.generalized_wrt(S(gamma))
phi1, inferred_type = infer(
S(recursion_env),
node.body,
is_top_level=False,
extensions=extensions,
initial_stack=declared_type.input,
)
# We want to check that the inferred outputs are subtypes of
# the declared outputs. Thus, inferred_type.output should be a subtype
# declared_type.output.
try:
inferred_type.output.constrain(declared_type.output)
except TypeError:
message = (
'declared function type {} is not compatible with '
'inferred type {}')
raise TypeError(
message.format(declared_type, inferred_type))
effect = declared_type
# we *mutate* the type environment
gamma[name] = effect.generalized_wrt(S(gamma))
elif isinstance(node,
concat.operators.GreaterThanOrEqualToWordNode):
a_type, b_type = current_effect.output[-2:]
try:
ge_type = a_type.get_type_of_attribute('__ge__')
if not isinstance(ge_type, PythonFunctionType):
raise TypeError(
'method __ge__ of type {} should be a Python function'
.format(ge_type))
_, current_subs = ge_type.select_overload([b_type])
except TypeError:
le_type = b_type.get_type_of_attribute('__le__')
if not isinstance(le_type, PythonFunctionType):
raise TypeError(
'method __le__ of type {} should be a Python function'
.format(le_type))
_, current_subs = le_type.select_overload([a_type])
current_subs, current_effect = (
current_subs,
StackEffect(
current_effect.input,
TypeSequence([*current_effect.output[:-2], bool_type]),
),
)
elif isinstance(
node,
(
concat.operators.IsWordNode,
concat.operators.AndWordNode,
concat.operators.OrWordNode,
concat.operators.EqualToWordNode,
),
):
# TODO: I should be more careful here, since at least __eq__ can be
# deleted, if I remember correctly.
if not isinstance(current_effect.output[-1],
IndividualType) or not isinstance(
current_effect.output[-2],
IndividualType):
raise StackMismatchError(
TypeSequence(current_effect.output),
TypeSequence([object_type, object_type]),
)
current_effect = StackEffect(
current_effect.input,
TypeSequence([*current_effect.output[:-2], bool_type]),
)
elif isinstance(node, concat.parse.NumberWordNode):
if isinstance(node.value, int):
current_effect = StackEffect(i, [*o, int_type])
else:
raise UnhandledNodeTypeError
elif isinstance(node, concat.parse.NameWordNode):
(i1, o1) = current_effect
if node.value not in current_subs(gamma):
raise NameError(node)
type_of_name = current_subs(gamma)[node.value].instantiate()
type_of_name = type_of_name.get_type_of_attribute('__call__')
if not isinstance(type_of_name, StackEffect):
raise UnhandledNodeTypeError(
'name {} of type {} (repr {!r})'.format(
node.value, type_of_name, type_of_name))
constraint_subs = o1.constrain_and_bind_supertype_variables(
type_of_name.input, set())
current_subs = constraint_subs(current_subs)
current_effect = current_subs(
StackEffect(i1, type_of_name.output))
elif isinstance(node, concat.parse.QuoteWordNode):
quotation = cast(concat.parse.QuoteWordNode, node)
# make sure any annotation matches the current stack
if quotation.input_stack_type is not None:
input_stack, _ = quotation.input_stack_type.to_type(gamma)
S = TypeSequence(o).constrain_and_bind_supertype_variables(
input_stack,
set(),
)(S)
else:
input_stack = TypeSequence(o)
S1, (i1, o1) = infer(
gamma,
[*quotation.children],
extensions=extensions,
source_dir=source_dir,
initial_stack=input_stack,
)
current_subs, current_effect = (
S1(S),
S1(StackEffect(i, o1)),
)
elif isinstance(node, concat.parse.StringWordNode):
current_subs, current_effect = (
S,
StackEffect(
current_effect.input,
[*current_effect.output, str_type],
),
)
elif isinstance(node, concat.parse.AttributeWordNode):
stack_top_type = o[-1]
out_types = o[:-1]
attr_function_type = stack_top_type.get_type_of_attribute(
node.value).instantiate()
if not isinstance(attr_function_type, StackEffect):
raise UnhandledNodeTypeError(
'attribute {} of type {} (repr {!r})'.format(
node.value, attr_function_type,
attr_function_type))
R = TypeSequence(
out_types).constrain_and_bind_supertype_variables(
attr_function_type.input,
set(),
)
current_subs, current_effect = (
R(S),
R(StackEffect(i, attr_function_type.output)),
)
elif isinstance(node, concat.parse.CastWordNode):
new_type, _ = node.type.to_type(gamma)
rest = current_effect.output[:-1]
current_effect = current_subs(
StackEffect(current_effect.input, [*rest, new_type]))
else:
raise UnhandledNodeTypeError(
"don't know how to handle '{}'".format(node))
except TypeError as e:
e.set_location_if_missing(node.location)
raise
return current_subs, current_effect
Mapping,
Optional,
Sequence,
Sized,
Union,
cast,
)
_stack_type_var = SequenceVariable()
_rest_var = SequenceVariable()
_rest_var_2 = SequenceVariable()
_rest_var_3 = SequenceVariable()
globals()['@@types'] = {
'to_str': StackEffect(
TypeSequence([_stack_type_var, object_type, object_type, object_type]),
TypeSequence([_stack_type_var, str_type]),
),
'to_py_function': ForAll(
[_rest_var, _rest_var_2, _rest_var_3],
StackEffect(
[_rest_var, StackEffect([_rest_var_2], [_rest_var_3])],
[_rest_var, py_function_type],
),
),
}
def to_py_function(stack: List[object], stash: List[object]) -> None:
func = cast(Callable[[List[object], List[object]], None], stack.pop())