gh-159672: Release trial identifiers with their arena

Skip immortal interning for replay-only identifiers and dotted names.
The parser can then release those strings after each trial.

Author: pablogsal

Lib/test/test_traceback.py

@@ -1951,6 +1951,30 @@ def test_keyword_trials_do_not_call_normalizer(self):
         normalize.assert_not_called()
         self.assertIsNone(caught.exception._metadata)
 
+    @cpython_only
+    @requires_subprocess()
+    def test_keyword_trials_do_not_intern_names(self):
+        assert_python_ok('-c', textwrap.dedent('''
+            import sys
+
+            def check(index, suffix):
+                source = (f"impor math\\nidentifier_after_typo_{index} = 1\\n"
+                          f"import module_after_typo_{index}.child" + suffix)
+                try:
+                    compile(source, "<test>", "exec")
+                except SyntaxError as error:
+                    assert (error._metadata is not None) == (not suffix)
+                else:
+                    raise AssertionError("expected SyntaxError")
+
+            for suffix in ("", "\\nother ="):
+                check(0, suffix)
+                before = sys.getunicodeinternedsize(_only_immortal=True)
+                for index in range(1, 20):
+                    check(index, suffix)
+                assert sys.getunicodeinternedsize(_only_immortal=True) == before
+        '''))
+
     @cpython_only
     def test_keyword_replay_source_limit(self):
         with tempfile.TemporaryDirectory() as script_dir:

Parser/action_helpers.c

@@ -128,8 +128,10 @@ _PyPegen_join_names_with_dot(Parser *p, expr_ty first_name, expr_ty second_name)
     if (!uni) {
         return NULL;
     }
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    _PyUnicode_InternImmortal(interp, &uni);
+    if (p->keyword_trial == NULL) {
+        PyInterpreterState *interp = _PyInterpreterState_GET();
+        _PyUnicode_InternImmortal(interp, &uni);
+    }
     if (_PyArena_AddPyObject(p->arena, uni) < 0) {
         Py_DECREF(uni);
         return NULL;

Parser/pegen.c

@@ -662,8 +662,11 @@ _PyPegen_new_identifier(Parser *p, const char *n)
             goto error;
         }
     }
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    _PyUnicode_InternImmortal(interp, &id);
+    // Trial-only identifiers must die with the trial arena.
+    if (p->keyword_trial == NULL) {
+        PyInterpreterState *interp = _PyInterpreterState_GET();
+        _PyUnicode_InternImmortal(interp, &id);
+    }
     if (_PyArena_AddPyObject(p->arena, id) < 0)
     {
         Py_DECREF(id);
@@ -687,7 +690,7 @@ get_cached_identifier(Parser *p, PyObject *bytes)
     }
     // Identifiers repeat constantly; a small span-keyed cache skips the
     // UTF-8 decode + intern for repeated occurrences. Keys point into
-    // arena-owned token bytes and values are arena-owned interned strings,
+    // arena-owned token bytes and values are arena-owned strings,
     // so borrowed references are valid for the lifetime of the parse
     // (including the second error pass, which reuses parser and arena).
     Py_ssize_t len = PyBytes_GET_SIZE(bytes);