Add test suite for operators typechecking (#17)

While doing so, fixed a few issues:
Fixed reference parsing when trying to resolve the empty pointers
(i.e. `""`). Note that dot references does not support such of format
as they are ambiguous.
Fixed `DataStack` crash when no initial data is provided.
Fixed some diagnostic messages in operators typechecking.
Fixed construction of some operators.

src/jsonlogic/operators/operators.py

@@ -9,7 +9,15 @@
 from jsonlogic.core import JSONLogicSyntaxError, Operator
 from jsonlogic.evaluation import EvaluationContext
 from jsonlogic.json_schema import as_json_schema, from_json_schema
-from jsonlogic.json_schema.types import AnyType, ArrayType, BinaryOp, BooleanType, JSONSchemaType, UnsupportedOperation
+from jsonlogic.json_schema.types import (
+    AnyType,
+    ArrayType,
+    BinaryOp,
+    BooleanType,
+    JSONSchemaType,
+    StringType,
+    UnsupportedOperation,
+)
 from jsonlogic.resolving import Unresolvable
 from jsonlogic.typing import OperatorArgument
 from jsonlogic.utils import UNSET, UnsetType
@@ -37,7 +45,11 @@ def from_expression(cls, operator: str, arguments: list[OperatorArgument]) -> Se
 
     def typecheck(self, context: TypecheckContext) -> JSONSchemaType:
         if isinstance(self.variable_path, Operator):
-            self.variable_path.typecheck(context)
+            var_type = self.variable_path.typecheck(context)
+            if not isinstance(var_type, StringType):
+                context.add_diagnostic(
+                    f"The first argument must be of type string, got {var_type.name}", "argument_type", self
+                )
             return AnyType()
 
         default_value_type: JSONSchemaType | None
@@ -51,15 +63,15 @@ def typecheck(self, context: TypecheckContext) -> JSONSchemaType:
         except Unresolvable:
             if default_value_type is None:
                 context.add_diagnostic(
-                    f"{self.variable_path} is unresolvable and no fallback value is provided",
+                    f"{self.variable_path!r} is unresolvable and no fallback value is provided",
                     "unresolvable_variable",
                     self,
                 )
                 return AnyType()
 
             # We emit a diagnostic but as a warning, as this will not fail at runtime
             context.add_diagnostic(
-                f"{self.variable_path} is unresolvable", "unresolvable_variable", self, type="warning"
+                f"{self.variable_path!r} is unresolvable", "unresolvable_variable", self, type="warning"
             )
             return default_value_type
         else:
@@ -114,15 +126,21 @@ class If(Operator):
 
     @classmethod
     def from_expression(cls, operator: str, arguments: list[OperatorArgument]) -> Self:
+        if len(arguments) <= 2:
+            raise JSONLogicSyntaxError(f"{operator!r} expects at least 3 arguments, got {len(arguments)}")
         if len(arguments) % 2 == 0:
             raise JSONLogicSyntaxError(f"{operator!r} expects an odd number of arguments, got {len(arguments)}")
         return cls(operator=operator, if_elses=list(zip(arguments[::2], arguments[1::2])), leading_else=arguments[-1])
 
     def typecheck(self, context: TypecheckContext) -> JSONSchemaType:
         for i, (cond, _) in enumerate(self.if_elses, start=1):
             cond_type = get_type(cond, context)
-            if not isinstance(cond_type, BooleanType):
-                context.add_diagnostic(f"Condition {i} should be a boolean", "argument_type", self)
+            try:
+                # TODO It might be that unary_op("bool") does not return
+                # `BooleanType`, altough it wouldn't make much sense
+                cond_type.unary_op("bool")
+            except UnsupportedOperation:
+                context.add_diagnostic(f"Condition {i} should support boolean evaluation", "argument_type", self)
 
         return functools.reduce(operator.or_, (get_type(rv, context) for _, rv in self.if_elses)) | get_type(
             self.leading_else, context
@@ -213,7 +231,7 @@ class Plus(Operator):
 
     @classmethod
     def from_expression(cls, operator: str, arguments: list[OperatorArgument]) -> Self:
-        # Having a unary + is a bit useless, but we might not error here in the future
+        # TODO Having a unary + is a bit useless, but we might not error here in the future
         if not len(arguments) >= 2:
             raise JSONLogicSyntaxError(f"{operator!r} expects at least two arguments, got {len(arguments)}")
         return cls(operator=operator, arguments=arguments)
@@ -287,15 +305,17 @@ class Map(Operator):
 
     @classmethod
     def from_expression(cls, operator: str, arguments: list[OperatorArgument]) -> Self:
-        if len(arguments) not in {1, 2}:
+        if len(arguments) != 2:
             raise JSONLogicSyntaxError(f"{operator!r} expects two arguments, got {len(arguments)}")
 
         return cls(operator=operator, vars=arguments[0], func=arguments[1])
 
     def typecheck(self, context: TypecheckContext) -> JSONSchemaType:
         vars_type = get_type(self.vars, context)
         if not isinstance(vars_type, ArrayType):
-            context.add_diagnostic(f"The first argument must be of type array, got {vars_type}", "argument_type", self)
+            context.add_diagnostic(
+                f"The first argument must be of type array, got {vars_type.name}", "argument_type", self
+            )
             return AnyType()
 
         vars_type = cast(ArrayType[JSONSchemaType], vars_type)

src/jsonlogic/resolving.py

@@ -90,18 +90,21 @@ class DotReferenceParser(BaseReferenceParser):
     def __call__(self, reference: str) -> tuple[ParsedReference, int]:
         reference, scope = self.parse_scope(reference)
 
-        return ParsedReference(reference, reference.split(".")), scope
+        return ParsedReference(reference, [] if reference == "" else reference.split(".")), scope
 
 
 class PointerReferenceParser(BaseReferenceParser):
     """A reference parser able to parse JSON Pointer references, as specified by :rfc:`6901`."""
 
     def __call__(self, reference: str) -> tuple[ParsedReference, int]:
         reference, scope = self.parse_scope(reference)
-        segments = [
-            segment.replace("~2", "@").replace("~1", "/").replace("~0", "~")
-            for segment in unquote(reference[1:]).split("/")
-        ]
+        if reference == "":
+            segments = []
+        else:
+            segments = [
+                segment.replace("~2", "@").replace("~1", "/").replace("~0", "~")
+                for segment in unquote(reference[1:]).split("/")
+            ]
 
         return ParsedReference(reference, segments), scope
 

src/jsonlogic/utils.py

@@ -26,8 +26,9 @@ def __repr__(self) -> str:
 
 class DataStack(Generic[DataT]):
     def __init__(self, root_data: DataT | UnsetType = UNSET) -> None:
+        self._stack = []
         if root_data is not UNSET:
-            self._stack = [root_data]
+            self._stack.append(root_data)
 
     @property
     def tail(self) -> DataT:

tests/operators/test_from_expression.py

@@ -46,6 +46,9 @@ def test_if() -> None:
     assert if_.if_elses == [(1, 2), (3, 4)]
     assert if_.leading_else == 5
 
+    with pytest.raises(JSONLogicSyntaxError, match="'if' expects at least 3 arguments, got 1"):
+        If.from_expression("if", [1])
+
     with pytest.raises(JSONLogicSyntaxError, match="'if' expects an odd number of arguments, got 6"):
         If.from_expression("if", [1, 2, 3, 4, 5, 6])
 

tests/operators/test_typecheck.py

@@ -0,0 +1,235 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, cast
+
+from jsonlogic._compat import Self
+from jsonlogic.core import JSONLogicExpression, Operator
+from jsonlogic.json_schema.types import (
+    AnyType,
+    ArrayType,
+    BooleanType,
+    DateType,
+    IntegerType,
+    JSONSchemaType,
+    NumberType,
+    StringType,
+    UnionType,
+)
+from jsonlogic.operators import operator_registry as base_operator_registry
+from jsonlogic.typechecking import TypecheckContext, typecheck
+from jsonlogic.typing import OperatorArgument
+
+
+@dataclass
+class ReturnsOp(Operator):
+    """A test operator returning the provided type during typechecking."""
+
+    return_type: JSONSchemaType
+
+    @classmethod
+    def from_expression(cls, operator: str, arguments: list[OperatorArgument]) -> Self:
+        assert len(arguments) == 1
+        return cls(
+            operator=operator,
+            return_type=arguments[0],  # type: ignore
+        )
+
+    def typecheck(self, context: TypecheckContext) -> JSONSchemaType:
+        return self.return_type
+
+
+operator_registry = base_operator_registry.copy(extend={"returns": ReturnsOp})
+
+
+def as_op(json_logic: dict[str, Any]) -> Operator:
+    expr = JSONLogicExpression.from_json(json_logic)
+    return cast(Operator, expr.as_operator_tree(operator_registry))
+
+
+def test_var_dynamic_variable_path() -> None:
+    op_no_diag = as_op({"var": {"returns": StringType()}})
+    rt, diagnostics = typecheck(op_no_diag, {})
+
+    assert rt == AnyType()
+    assert diagnostics == []
+
+    op_diag = as_op({"var": {"returns": BooleanType()}})
+    rt, diagnostics = typecheck(op_diag, {})
+
+    assert rt == AnyType()
+    diag = diagnostics[0]
+    assert diag.category == "argument_type"
+    assert diag.message == "The first argument must be of type string, got boolean"
+
+
+def test_var_unresolvable_no_default() -> None:
+    op = as_op({"var": "/some_var"})
+    rt, diagnostics = typecheck(op, {})
+
+    assert rt == AnyType()
+    diag = diagnostics[0]
+
+    assert diag.category == "unresolvable_variable"
+    assert diag.message == "'/some_var' is unresolvable and no fallback value is provided"
+
+
+def test_var_unresolvable_default() -> None:
+    op = as_op({"var": ["/some_var", 1]})
+    rt, diagnostics = typecheck(op, {})
+
+    assert rt == IntegerType()
+    diag = diagnostics[0]
+
+    assert diag.category == "unresolvable_variable"
+    assert diag.message == "'/some_var' is unresolvable"
+    assert diag.type == "warning"
+
+
+def test_var_resolvable_no_default() -> None:
+    op = as_op({"var": "/some_var"})
+    rt, diagnostics = typecheck(op, {"type": "object", "properties": {"some_var": {"type": "string"}}})
+
+    assert rt == StringType()
+    assert diagnostics == []
+
+
+def test_var_resolvable_default() -> None:
+    op = as_op({"var": ["/some_var", 1]})
+    rt, diagnostics = typecheck(op, {"type": "object", "properties": {"some_var": {"type": "string"}}})
+
+    assert rt == UnionType(StringType(), IntegerType())
+    assert diagnostics == []
+
+
+def test_equality_op() -> None:
+    # Note: this test is relevant for the `Equal` and `NotEqual` operator classes.
+    op_1 = as_op({"==": [1, "test"]})
+    rt, diagnostics = typecheck(op_1, {})
+
+    assert rt == BooleanType()
+    assert diagnostics == []
+
+    op_2 = as_op({"==": [{"var": "a"}, {"var": "b"}]})
+    rt, diagnostics = typecheck(op_2, {})
+
+    assert rt == BooleanType()
+    # Should contain diagnostics from the var op:
+    assert len(diagnostics) == 2
+
+
+def test_if_bad_condition() -> None:
+    # `DateType` does not support boolean evaluation
+    returns_date = {"returns": DateType()}
+    op = as_op({"if": [returns_date, "some_value", "other_value"]})
+    _, diagnostics = typecheck(op, {})
+    diag = diagnostics[0]
+
+    assert diag.category == "argument_type"
+    assert diag.message == "Condition 1 should support boolean evaluation"
+
+    op = as_op({"if": [returns_date, "some_value", returns_date, "other_value", "another_value"]})
+    _, diagnostics = typecheck(op, {})
+    diag_1, diag_2 = diagnostics
+
+    assert diag_1.message == "Condition 1 should support boolean evaluation"
+    assert diag_2.message == "Condition 2 should support boolean evaluation"
+
+
+def test_if() -> None:
+    op = as_op({"if": [1, "string_val", 2, 2.0, "other_string"]})
+    rt, diagnostics = typecheck(op, {})
+
+    assert rt == UnionType(StringType(), NumberType())
+    assert diagnostics == []
+
+
+def test_binary_op() -> None:
+    # Note: this test is relevant for all the binary operator classes.
+    op = as_op({">": [2, {"returns": NumberType()}]})
+    rt, diagnostics = typecheck(op, {})
+
+    assert rt == BooleanType()
+    assert diagnostics == []
+
+    op_fail = as_op({">=": [2, {"returns": DateType()}]})
+    rt, diagnostics = typecheck(op_fail, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "operator"
+    assert diag.message == 'Operator ">=" not supported for types integer and date'
+
+
+def test_plus() -> None:
+    op_two_operands = as_op({"+": ["a", "b"]})
+    rt, diagnostics = typecheck(op_two_operands, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "operator"
+    assert diag.message == 'Operator "+" not supported for types string and string'
+
+    op_three_operands = as_op({"+": ["a", 1, 2]})
+    rt, diagnostics = typecheck(op_three_operands, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "operator"
+    assert diag.message == 'Operator "+" not supported for types string (argument 1) and integer (argument 2)'
+
+    op_ok = as_op({"+": [1, 2, 3, 4]})
+    rt, diagnostics = typecheck(op_ok, {})
+
+    assert rt == IntegerType()
+    assert diagnostics == []
+
+
+def test_minus() -> None:
+    op_unary = as_op({"-": {"returns": DateType()}})
+    rt, diagnostics = typecheck(op_unary, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "operator"
+    assert diag.message == 'Operator "-" not supported for type date'
+
+    op_binary = as_op({"-": [{"returns": DateType()}, {"returns": IntegerType()}]})
+    rt, diagnostics = typecheck(op_binary, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "operator"
+    assert diag.message == 'Operator "-" not supported for type date and integer'
+
+    op_ok = as_op({"-": [1, 1]})
+    rt, diagnostics = typecheck(op_ok, {})
+
+    assert rt == IntegerType()
+    assert diagnostics == []
+
+
+def test_map() -> None:
+    op_bad_vars = as_op({"map": [1, "irrelevant"]})
+    rt, diagnostics = typecheck(op_bad_vars, {})
+    diag = diagnostics[0]
+
+    assert rt == AnyType()
+    assert diag.category == "argument_type"
+    assert diag.message == "The first argument must be of type array, got integer"
+
+    op = as_op({"map": [[1, 2], {"+": [{"var": ""}, 2.0]}]})
+    rt, diagnostics = typecheck(op, {})
+
+    assert rt == ArrayType(NumberType())
+    assert diagnostics == []
+
+
+def test_map_root_reference() -> None:
+    # The `/@1` reference should resolve to the "" attribute of the top level schema,
+    # meaning the variables of the `map` operators are meaningless.
+    op = as_op({"map": [["some", "strings"], {"+": [{"var": "/@1"}, 2.0]}]})
+    rt, diagnostics = typecheck(op, {"type": "object", "properties": {"": {"type": "integer"}}})
+
+    assert rt == ArrayType(NumberType())
+    assert diagnostics == []