[3.9] bpo-41631: _ast module uses again a global state (GH-21961) (GH-22258)

Partially revert commit ac46eb4ad6662cf6d771b20d8963658b2186c48c:
"bpo-38113: Update the Python-ast.c generator to PEP384 (gh-15957)".

Using a module state per module instance is causing subtle practical
problems.

For example, the Mercurial project replaces the __import__() function
to implement lazy import, whereas Python expected that "import _ast"
always return a fully initialized _ast module.

Add _PyAST_Fini() to clear the state at exit.

The _ast module has no state (set _astmodule.m_size to 0). Remove
astmodule_traverse(), astmodule_clear() and astmodule_free()
functions..
(cherry picked from commit e5fbe0cbd4be99ced5f000ad382208ad2a561c90)

Co-authored-by: Victor Stinner <vstinner@python.org>

2 files changed, 106 insertions, 15 deletions

diff --git a/Lib/ast.py b/Lib/ast.py
index 65ebd0100d..d860917f4d 100644
--- a/Lib/ast.py
+++ b/Lib/ast.py
@@ -497,18 +497,20 @@ class NodeTransformer(NodeVisitor):
         return node
 
 
-# The following code is for backward compatibility.
-# It will be removed in future.
+# If the ast module is loaded more than once, only add deprecated methods once
+if not hasattr(Constant, 'n'):
+    # The following code is for backward compatibility.
+    # It will be removed in future.
 
-def _getter(self):
-    """Deprecated. Use value instead."""
-    return self.value
+    def _getter(self):
+        """Deprecated. Use value instead."""
+        return self.value
 
-def _setter(self, value):
-    self.value = value
+    def _setter(self, value):
+        self.value = value
 
-Constant.n = property(_getter, _setter)
-Constant.s = property(_getter, _setter)
+    Constant.n = property(_getter, _setter)
+    Constant.s = property(_getter, _setter)
 
 class _ABC(type):
 
@@ -600,14 +602,19 @@ class ExtSlice(slice):
     def __new__(cls, dims=(), **kwargs):
         return Tuple(list(dims), Load(), **kwargs)
 
-def _dims_getter(self):
-    """Deprecated. Use elts instead."""
-    return self.elts
+# If the ast module is loaded more than once, only add deprecated methods once
+if not hasattr(Tuple, 'dims'):
+    # The following code is for backward compatibility.
+    # It will be removed in future.
 
-def _dims_setter(self, value):
-    self.elts = value
+    def _dims_getter(self):
+        """Deprecated. Use elts instead."""
+        return self.elts
 
-Tuple.dims = property(_dims_getter, _dims_setter)
+    def _dims_setter(self, value):
+        self.elts = value
+
+    Tuple.dims = property(_dims_getter, _dims_setter)
 
 class Suite(mod):
     """Deprecated AST node class.  Unused in Python 3."""
diff --git a/Lib/test/test_ast.py b/Lib/test/test_ast.py
index f5aef61ec6..5f57ce8724 100644
--- a/Lib/test/test_ast.py
+++ b/Lib/test/test_ast.py
@@ -1,7 +1,9 @@
 import ast
+import builtins
 import dis
 import os
 import sys
+import types
 import unittest
 import warnings
 import weakref
@@ -1945,6 +1947,88 @@ class NodeVisitorTests(unittest.TestCase):
         ])
 
 
+@support.cpython_only
+class ModuleStateTests(unittest.TestCase):
+    # bpo-41194, bpo-41261, bpo-41631: The _ast module uses a global state.
+
+    def check_ast_module(self):
+        # Check that the _ast module still works as expected
+        code = 'x + 1'
+        filename = '<string>'
+        mode = 'eval'
+
+        # Create _ast.AST subclasses instances
+        ast_tree = compile(code, filename, mode, flags=ast.PyCF_ONLY_AST)
+
+        # Call PyAST_Check()
+        code = compile(ast_tree, filename, mode)
+        self.assertIsInstance(code, types.CodeType)
+
+    def test_reload_module(self):
+        # bpo-41194: Importing the _ast module twice must not crash.
+        with support.swap_item(sys.modules, '_ast', None):
+            del sys.modules['_ast']
+            import _ast as ast1
+
+            del sys.modules['_ast']
+            import _ast as ast2
+
+            self.check_ast_module()
+
+        # Unloading the two _ast module instances must not crash.
+        del ast1
+        del ast2
+        support.gc_collect()
+
+        self.check_ast_module()
+
+    def test_sys_modules(self):
+        # bpo-41631: Test reproducing a Mercurial crash when PyAST_Check()
+        # imported the _ast module internally.
+        lazy_mod = object()
+
+        def my_import(name, *args, **kw):
+            sys.modules[name] = lazy_mod
+            return lazy_mod
+
+        with support.swap_item(sys.modules, '_ast', None):
+            del sys.modules['_ast']
+
+            with support.swap_attr(builtins, '__import__', my_import):
+                # Test that compile() does not import the _ast module
+                self.check_ast_module()
+                self.assertNotIn('_ast', sys.modules)
+
+                # Sanity check of the test itself
+                import _ast
+                self.assertIs(_ast, lazy_mod)
+
+    def test_subinterpreter(self):
+        # bpo-41631: Importing and using the _ast module in a subinterpreter
+        # must not crash.
+        code = dedent('''
+            import _ast
+            import ast
+            import gc
+            import sys
+            import types
+
+            # Create _ast.AST subclasses instances and call PyAST_Check()
+            ast_tree = compile('x+1', '<string>', 'eval',
+                               flags=ast.PyCF_ONLY_AST)
+            code = compile(ast_tree, 'string', 'eval')
+            if not isinstance(code, types.CodeType):
+                raise AssertionError
+
+            # Unloading the _ast module must not crash.
+            del ast, _ast
+            del sys.modules['ast'], sys.modules['_ast']
+            gc.collect()
+        ''')
+        res = support.run_in_subinterp(code)
+        self.assertEqual(res, 0)
+
+
 def main():
     if __name__ != '__main__':
         return