Il 16/07/2013 21:11, Gwenaël Casaccio ha scritto: >> To catch more bugs like this, we need to assert that add_literal is >> called in the outermost incubator context. We can add an >> inc_context_depth() function (which really can just return >> INC_SAVE_POINTER()), store the value in _gst_compile_method and compare >> it in add_literal. Let me try to do a patch. > > Yes but you capture only one kind of bugs tracking allocation with a > "global" incubator is easier.
There is a global incubator, that's _gst_register_oop and _gst_register_oop_array. INC_SAVE_POINTER/INC_RESTORE_POINTER is explicitly for stack-based operation. add_literal is at the bottom of the stack because the objects it incubates are for the outermost object (the CompiledMethod). So the invariant makes sense, and the fragility might only be apparent (i.e. it actually helps finding bugs). That said, I'm glad to evaluate a patch to remove incubator usage from add_literal, and instead use _gst_register_oop_array. Paolo _______________________________________________ help-smalltalk mailing list help-smalltalk@gnu.org https://lists.gnu.org/mailman/listinfo/help-smalltalk
