Update comments and docstrings related to binder

mypy/binder.py

@@ -41,12 +41,22 @@ class Frame:
     """A Frame represents a specific point in the execution of a program.
     It carries information about the current types of expressions at
     that point, arising either from assignments to those expressions
-    or the result of isinstance checks. It also records whether it is
-    possible to reach that point at all.
+    or the result of isinstance checks and other type narrowing
+    operations. It also records whether it is possible to reach that
+    point at all.
+
+    We add a new frame wherenever there is a new scope or control flow
+    branching.
 
     This information is not copied into a new Frame when it is pushed
     onto the stack, so a given Frame only has information about types
     that were assigned in that frame.
+
+    Expressions are stored in dicts using 'literal hashes' as keys (type
+    "Key"). These are hashable values derived from expression AST nodes
+    (only those that can be narrowed). literal_hash(expr) is used to
+    calculate the hashes. Note that this isn't directly related to literal
+    types -- the concept predates literal types.
     """
 
     def __init__(self, id: int, conditional_frame: bool = False) -> None:
@@ -66,29 +76,29 @@ def __repr__(self) -> str:
 class ConditionalTypeBinder:
     """Keep track of conditional types of variables.
 
-    NB: Variables are tracked by literal expression, so it is possible
-    to confuse the binder; for example,
-
-    ```
-    class A:
-        a: Union[int, str] = None
-    x = A()
-    lst = [x]
-    reveal_type(x.a)      # Union[int, str]
-    x.a = 1
-    reveal_type(x.a)      # int
-    reveal_type(lst[0].a) # Union[int, str]
-    lst[0].a = 'a'
-    reveal_type(x.a)      # int
-    reveal_type(lst[0].a) # str
-    ```
+    NB: Variables are tracked by literal hashes of expressions, so it is
+    possible to confuse the binder when there is aliasing. Example:
+
+        class A:
+            a: int | str
+
+        x = A()
+        lst = [x]
+        reveal_type(x.a)      # int | str
+        x.a = 1
+        reveal_type(x.a)      # int
+        reveal_type(lst[0].a) # int | str
+        lst[0].a = 'a'
+        reveal_type(x.a)      # int
+        reveal_type(lst[0].a) # str
     """
 
     # Stored assignments for situations with tuple/list lvalue and rvalue of union type.
     # This maps an expression to a list of bound types for every item in the union type.
     type_assignments: Assigns | None = None
 
     def __init__(self) -> None:
+        # Each frame gets an increasing, distinct id.
         self.next_id = 1
 
         # The stack of frames currently used.  These map
@@ -116,6 +126,7 @@ def __init__(self) -> None:
         # Whether the last pop changed the newly top frame on exit
         self.last_pop_changed = False
 
+        # These are used to track control flow in try statements and loops.
         self.try_frames: set[int] = set()
         self.break_frames: list[int] = []
         self.continue_frames: list[int] = []
@@ -488,6 +499,11 @@ def top_frame_context(self) -> Iterator[Frame]:
 
 
 def get_declaration(expr: BindableExpression) -> Type | None:
+    """Get the declared or inferred type of a RefExpr expression.
+
+    Return None if there is no type or the expression is not a RefExpr.
+    This can return None if the type hasn't been inferred yet.
+    """
     if isinstance(expr, RefExpr):
         if isinstance(expr.node, Var):
             type = expr.node.type

mypy/literals.py

@@ -96,7 +96,27 @@
 #   of an index expression, or the operands of an operator expression).
 
 
+Key: _TypeAlias = tuple[Any, ...]
+
+
+def literal_hash(e: Expression) -> Key | None:
+    """Generate a hashable, (mostly) opaque key for expressions supported by the binder.
+
+    These allow using expressions as dictionary keys based on structural/value
+    matching (instead of based on expression identity).
+
+    Return None if the expression type is not supported (it cannot be narrowed).
+
+    See the comment above for more information.
+
+    NOTE: This is not directly related to literal types.
+    """
+    return e.accept(_hasher)
+
+
 def literal(e: Expression) -> int:
+    """Return the literal kind for an expression."""
+
     if isinstance(e, ComparisonExpr):
         return min(literal(o) for o in e.operands)
 
@@ -129,17 +149,10 @@ def literal(e: Expression) -> int:
     return LITERAL_NO
 
 
-Key: _TypeAlias = tuple[Any, ...]
-
-
 def subkeys(key: Key) -> Iterable[Key]:
     return [elt for elt in key if isinstance(elt, tuple)]
 
 
-def literal_hash(e: Expression) -> Key | None:
-    return e.accept(_hasher)
-
-
 def extract_var_from_literal_hash(key: Key) -> Var | None:
     """If key refers to a Var node, return it.
 

mypy/nodes.py

@@ -89,9 +89,14 @@ def set_line(
 REVEAL_TYPE: Final = 0
 REVEAL_LOCALS: Final = 1
 
-LITERAL_YES: Final = 2
-LITERAL_TYPE: Final = 1
-LITERAL_NO: Final = 0
+# Kinds of 'literal' expressions.
+#
+# Use the function mypy.literals.literal to calculate these.
+#
+# TODO: Can we make these less confusing?
+LITERAL_YES: Final = 2  # Value of expression known statically
+LITERAL_TYPE: Final = 1  # Type of expression can be narrowed (e.g. variable reference)
+LITERAL_NO: Final = 0  # None of the above
 
 node_kinds: Final = {LDEF: "Ldef", GDEF: "Gdef", MDEF: "Mdef", UNBOUND_IMPORTED: "UnboundImported"}
 inverse_node_kinds: Final = {_kind: _name for _name, _kind in node_kinds.items()}