Improve robustness on free threaded builds and add a TODO

Author: JukkaL

mypyc/codegen/emitclass.py

@@ -855,7 +855,6 @@ def emit_reuse_dealloc(cl: ClassIR, emitter: Emitter) -> None:
     """
     prefix = cl.name_prefix(emitter.names)
     emitter.emit_line(f"if ({prefix}_free_instance == NULL) {{")
-    emitter.emit_line(f"{prefix}_free_instance = self;")
 
     # Clear attributes and free referenced objects.
 
@@ -865,6 +864,11 @@ def emit_reuse_dealloc(cl: ClassIR, emitter: Emitter) -> None:
         for attr, rtype in base.attributes.items():
             emitter.emit_reuse_clear(f"self->{emitter.attr(attr)}", rtype)
 
+    # TODO: Insert a memory barrier on free-threaded builds? This appears not to be
+    #       needed on x86-64 because of the memory model.
+
+    emitter.emit_line(f"{prefix}_free_instance = self;")
+
     emitter.emit_line("return;")
     emitter.emit_line("}")