>> One possible solution is to somehow redirect every __builtins__ to a >> function that returns a different __builtins__ dictionary for each thread >> (such a function already exists).
>How exactly does the code reference it? If they're simply referring to >the name __builtins__ at module level, you ought to be able to import >the module, then assign some_module.__builtins__ to your thread-local >object, then call code in it as normal. >An interesting problem, and one where monkeypatching is, imho, justified. >ChrisA Hello, and thanks for your answer. Unfortunately, replacing __builtins__ at import time won't do, because external modules (that is, .py) get imported only once when they are accessed by the first thread, which includes (of course) setting up of __dict__ and __builtins__. When a second thread later accesses this module, it has the same variables in __builtins__ that were added by the same module in first thread And if the second thread then changes the values, I can see these same changes in the first thread. -> The problem is that __builtins__ are global, not thread-safe. The only solution I can see is therfor redirecting __builtins__ to a function which returns a different dictionary for each thread, e.g. by intercepting __builtins__-calls with __readattr__. To do this, I would need my own class to define __readattr__ in since (as far as I know) I can't define __readattr__ in a module, and I can't change metaclass <module> of course. I really don't know how to get around this problem...
-- http://mail.python.org/mailman/listinfo/python-list
