Getting namespaces right when parsing/executing Python ASTs

[Orestis Markou]

Have you tried passing in empty dicts for globals and locals? I think
that the defaults will be the *current* globals and locals, and then
of course your namespace is broken...

On Tue, Oct 7, 2008 at 1:26 PM, Gordon Fraser <gfraser79 at gmail.com> wrote:
> Hi,
>
> I'm trying to parse Python code to an AST, apply some changes to the AST
> and then compile and run the AST, but I'm running into problems when
> trying to evaluate/execute the resulting code object. It seems that the
> global namespace differs depending on where I call parse and eval/exec.
>
> The following code parses a file, compiles and then evaluates the AST.
> If I call Python directly on this code, then it works:
>
>        import sys, parser
>        ast = parser.suite(open(sys.argv[1]).read())
>        code = ast.compile()
>        exec(code)
>
> ...and it also works this way with Python2.6:
>
>    ast = compile(open(sys.argv[1]).read(), "<AST>",
> 'exec',_ast.PyCF_ONLY_AST)
>    code = compile(ast, "<AST", "exec")
>    exec(code)
>
> However, if I include that snippet in a different scope (some function
> or class), then the namespace that the code object will have differs -
> it seems the symbols defined in the AST are not included when executing
> the code. For example:
>
>        import sys,parser
>
>        def main():
>            ast = parser.suite(open(sys.argv[1]).read())
>            code = ast.compile()
>            exec(code)
>
>        if __name__ == "__main__":
>          main()
>
> In particular this is a problem if I'm parsing a module with several
> functions - none of these functions actually ends up in the scope of the
> code object (same behavior with Python2.6 and the PyCF_ONLY_AST
> version).
>
> The function "exec" takes parameters for globals and locals, but I have
> no idea where to get these dictionaries from the "parser" module. My
> guess is that I am misunderstanding something about how Python treats
> namespaces. Can anyone help me here?
>
> Thanks,
> Gordon
>
>
> --
> http://mail.python.org/mailman/listinfo/python-list
>



-- 
orestis at orestis.gr
http://orestis.gr