Scopes work? I think – handles nested decls in functions

trans/tests/builtins_test.py

@@ -7,11 +7,16 @@ test = (2 + 3) * 4
 glob = 5
 
 def g():
+    a = 8
     if True:
+        b = 9
         if True:
+            c = 10
             if True:
-                x = 5
-    print(x)
+                d = a + b + c
+    if True:
+        e = d + 1
+    print(e)
 
 def f(x):
     return x + 1

trans/transpiler/__init__.py

@@ -146,8 +146,7 @@ class PrecedenceContext:
         self.op = op
 
     def __enter__(self):
-        if self.visitor.precedence[-1:] != [self.op]:
-            self.visitor.precedence.append(self.op)
+        self.visitor.precedence.append(self.op)
 
     def __exit__(self, exc_type, exc_val, exc_tb):
         self.visitor.precedence.pop()
@@ -290,18 +289,82 @@ class ExpressionVisitor(NodeVisitor):
             yield " : "
             yield from self.visit(node.orelse)
 
+@dataclass
+class VarDecl:
+    kind: VarKind
+    val: Optional[str]
 
 @dataclass
 class Scope:
     parent: Optional["Scope"] = None
-    vars: Dict[str, VarKind] = field(default_factory=dict)
+    is_function: bool = False
+    vars: Dict[str, VarDecl] = field(default_factory=dict)
 
-    def is_global(self):
+    def is_global(self) -> bool:
+        """
+        Determines whether this scope is the global scope. The global scope is the only scope to have no parent.
+        """
         return self.parent is None
 
     def exists(self, name: str) -> bool:
+        """
+        Determines whether a variable exists in the current scope or any parent scope.
+        """
         return name in self.vars or (self.parent is not None and self.parent.exists(name))
 
+    def exists_local(self, name: str) -> bool:
+        """
+        Determines whether a variable exists in the current function or global scope.
+        The check does not cross function boundaries; i.e. global variables are not taken into account from inside
+        functions.
+        """
+        return name in self.vars or (not self.is_function and self.parent is not None and self.parent.exists_local(name))
+
+    def child(self) -> "Scope":
+        """
+        Creates a child scope with a new variable dictionary.
+        This is used for first-level elements of a function.
+        """
+        return Scope(self, False, {})
+
+    def child_share(self) -> "Scope":
+        """
+        Creates a child scope sharing the variable dictionary with the parent scope.
+        This is used for Python blocks, which share the variable scope with their parent block.
+        """
+        return Scope(self, False, self.vars)
+
+    def function(self, **kwargs) -> "Scope":
+        """
+        Creates a function scope.
+        """
+        return Scope(self, True, **kwargs)
+
+    def is_root(self) -> Optional[Dict[str, VarDecl]]:
+        """
+        Determines whether this scope is a root scope.
+
+        A root scope is either the global scope, or the first inner scope of a function.
+        Variable declarations in the generated code only ever appear in root scopes.
+        :return: `None` if this scope is not a root scope, otherwise the variable dictionary of the root scope.
+        """
+        if self.parent is None:
+            return self.vars
+        if self.parent.is_function:
+            return self.parent.vars
+        return None
+
+    def declare(self, name: str, val: Optional[str] = None) -> Optional[str]:
+        if self.exists_local(name):
+            # If the variable already exists in the current function or global scope, we don't need to declare it again.
+            # This is simply an assignment.
+            return None
+        vdict, prefix = self.vars, ""
+        if (root_vars := self.is_root()) is not None:
+            vdict, prefix = root_vars, "auto "  # Root scope declarations can use `auto`.
+        vdict[name] = VarDecl(VarKind.LOCAL, val)
+        return prefix
+
 
 # noinspection PyPep8Naming
 class BlockVisitor(NodeVisitor):
@@ -340,44 +403,94 @@ class BlockVisitor(NodeVisitor):
         yield f"auto {node.name}"
         yield args
         yield "{"
-        inner = BlockVisitor(Scope(self._scope, vars={arg.arg: VarKind.LOCAL for arg in node.args.args}))
+        inner = BlockVisitor(self._scope.function())
         for child in node.body:
-            yield from inner.visit(child)
+            # Python uses module- and function- level scoping. Blocks, like conditionals and loops, do not form scopes
+            # on their own. Variables are still accessible in the remainder of the parent function or in the global
+            # scope if outside a function.
+            # This is different from C++, where scope is tied to any code block. To emulate this behavior, we need to
+            # declare all variables in the first inner scope of a function.
+            # For example,
+            # ```py
+            # def f():
+            #     if True:
+            #         x = 1
+            #     print(x)
+            # ```
+            # is translated to
+            # ```cpp
+            # auto f() {
+            #     decltype(1) x;
+            #     if (true) {
+            #         x = 1;
+            #     }
+            #     print(x);
+            # }
+            # ```
+            # `decltype` allows for proper typing (`auto` can't be used for variables with values later assigned, since
+            # this would require real type inference, akin to what Rust does).
+            # This is only done, though, for *nested* blocks of a function. Root-level variables are declared with
+            # `auto`:
+            # ```py
+            # x = 1
+            # def f():
+            #     y = 2
+            # ```
+            # is translated to
+            # ```cpp
+            # auto x = 1;
+            # auto f() {
+            #     auto y = 2;
+            # }
+            # ```
+            child_visitor = BlockVisitor(inner._scope.child())
+
+            # We need to do this in two-passes. This unfortunately breaks our nice generator state-machine architecture.
+            # Fair enough.
+            [*child_code] = child_visitor.visit(child)
+
+            # Hoist inner variables to the root scope.
+            for var, decl in child_visitor._scope.vars.items():
+                if decl.kind == VarKind.LOCAL:  # Nested declarations become `decltype` declarations.
+                    yield f"decltype({decl.val}) {var};"
+                elif decl.kind in (VarKind.GLOBAL, VarKind.NONLOCAL):  # `global` and `nonlocal` just get hoisted as-is.
+                    inner._scope.vars[var] = decl
+            yield from child_code  # Yeet back the child node code.
         yield "}"
 
     def visit_Global(self, node: ast.Global) -> Iterable[str]:
         for name in map(self.fix_name, node.names):
-            self._scope.vars[name] = VarKind.GLOBAL
+            self._scope.vars[name] = VarDecl(VarKind.GLOBAL, None)
         yield ""
 
     def visit_Nonlocal(self, node: ast.Nonlocal) -> Iterable[str]:
         for name in map(self.fix_name, node.names):
-            self._scope.vars[name] = VarKind.NONLOCAL
+            self._scope.vars[name] = VarDecl(VarKind.NONLOCAL, None)
         yield ""
 
     def visit_If(self, node: ast.If) -> Iterable[str]:
         if not node.orelse and compare_ast(node.test, ast.parse('__name__ == "__main__"', mode="eval").body):
-            yield "int main() {"
-            for child in node.body:
-                yield from self.visit(child)
-            yield "}"
+            # Special case handling for Python's interesting way of defining an entry point.
+            # I mean, it's not *that* bad, it's just an attempt at retrofitting an "entry point" logic in a scripting
+            # language that, by essence, uses "the start of the file" as the implicit entry point, since files are
+            # read and executed line-by-line, contrary to usual structured languages that mark a distinction between
+            # declarations (functions, classes, modules, ...) and code.
+            # Also, for nitpickers, the C++ standard explicitly allows for omitting a `return` statement in the `main`.
+            # 0 is returned by default.
+            yield "int main()"
+            yield from self.emit_block(node.body)
             return
 
         yield "if ("
         yield from ExpressionVisitor().visit(node.test)
-        yield ") {"
-        for child in node.body:
-            yield from self.visit(child)
-        yield "}"
+        yield ")"
+        yield from self.emit_block(node.body)
         if node.orelse:
             yield "else "
             if isinstance(node.orelse, ast.If):
                 yield from self.visit(node.orelse)
             else:
-                yield "{"
-                for child in node.orelse:
-                    yield from self.visit(child)
-                yield "}"
+                yield from self.emit_block(node.orelse)
 
     def visit_Return(self, node: ast.Return) -> Iterable[str]:
         yield "return "
@@ -388,21 +501,19 @@ class BlockVisitor(NodeVisitor):
     def visit_While(self, node: ast.While) -> Iterable[str]:
         yield "while ("
         yield from ExpressionVisitor().visit(node.test)
-        yield ") {"
-        for child in node.body:
-            yield from self.visit(child)
-        yield "}"
+        yield ")"
+        yield from self.emit_block(node.body)
         if node.orelse:
             raise NotImplementedError(node, "orelse")
 
-    def visit_lvalue(self, lvalue: ast.expr) -> Iterable[str]:
+    def visit_lvalue(self, lvalue: ast.expr, val: Optional[ast.AST] = None) -> Iterable[str]:
         if isinstance(lvalue, ast.Tuple):
             yield f"std::tie({', '.join(flatmap(ExpressionVisitor().visit, lvalue.elts))})"
         elif isinstance(lvalue, ast.Name):
             name = self.fix_name(lvalue.id)
-            if name not in self._scope.vars:
-                self._scope.vars[name] = name
-                yield "auto "
+            #if name not in self._scope.vars:
+            if not self._scope.exists_local(name):
+                yield self._scope.declare(name, " ".join(ExpressionVisitor().visit(val)) if val else None)
             yield name
         elif isinstance(lvalue, ast.Subscript):
             yield from ExpressionVisitor().visit(lvalue)
@@ -412,7 +523,7 @@ class BlockVisitor(NodeVisitor):
     def visit_Assign(self, node: ast.Assign) -> Iterable[str]:
         if len(node.targets) != 1:
             raise NotImplementedError(node)
-        yield from self.visit_lvalue(node.targets[0])
+        yield from self.visit_lvalue(node.targets[0], node.value)
         yield " = "
         yield from ExpressionVisitor().visit(node.value)
         yield ";"
@@ -420,7 +531,7 @@ class BlockVisitor(NodeVisitor):
     def visit_AnnAssign(self, node: ast.AnnAssign) -> Iterable[str]:
         if node.value is None:
             raise NotImplementedError(node, "empty value")
-        yield from self.visit_lvalue(node.target)
+        yield from self.visit_lvalue(node.target, node.value)
         yield " = "
         yield from ExpressionVisitor().visit(node.value)
         yield ";"
@@ -436,9 +547,20 @@ class BlockVisitor(NodeVisitor):
             raise NotImplementedError(node)
         yield f"for (auto {node.target.id} : "
         yield from ExpressionVisitor().visit(node.iter)
-        yield ") {"
-        for child in node.body:
-            yield from self.visit(child)
-        yield "}"
+        yield ")"
+        yield from self.emit_block(node.body)
         if node.orelse:
             raise NotImplementedError(node, "orelse")
+
+    def block(self) -> "BlockVisitor":
+        # See the comments in visit_FunctionDef.
+        # A Python code block does not introduce a new scope, so we create a new `Scope` object that shares the same
+        # variables as the parent scope.
+        return BlockVisitor(self._scope.child_share())
+
+    def emit_block(self, items: List[ast.stmt]) -> Iterable[str]:
+        yield "{"
+        block = self.block()
+        for child in items:
+            yield from block.visit(child)
+        yield "}"