Steven D'Aprano wrote in news:016abfa1$0$6988$c3e8...@news.astraweb.com in comp.lang.python:
> On Tue, 30 Dec 2008 15:35:28 -0600, Rob Williscroft wrote: > >> Stef Mientki wrote in news:mailman.6399.1230668197.3487.python- >> l...@python.org in comp.lang.python: >> >>>>> And, by the way, exec is a *statement*, not a function! >>>>> >>> exec ( Init_Code, PG.P_Globals ) >>> >>> I've really doubt that this is a statement, unless I don't understand >>> what a statement is. >>>>> >>>>> >>> >> In python 2.x the above is a statement that is passed a tuple: >> >> http://docs.python.org/reference/simple_stmts.html#exec > > > The documentation doesn't say anything about it accepting a tuple as an > argument. The tuple argument works in both 2.5 and 2.6. Curious. > My mistake, it is also behaving as a function: http://docs.python.org/dev/3.0/whatsnew/3.0.html#removed-syntax Removed keyword: exec() is no longer a keyword; it remains as a function. (Fortunately the function syntax was also accepted in 2.x.) Also ... Though that was the only documentation of it I found in a brief web search. > I was also surprised by this behaviour: > >>>> g, l = {}, {} # no globals, no locals >>>> exec "x = 1" in g, l >>>> l > {'x': 1} >>>> g.keys() > ['__builtins__'] > > I see *now* that this is documented: > > "...the current implementation MAY add a reference to the dictionary of > the built-in module __builtin__ under the key __builtins__ (!)." > [emphasis added] > > but it's still rather disconcerting. That means that: > > exec "some potentially dangerous code" in {}, {} > > isn't as safe as I thought it was. AIUI it isn't meant to be safe, it provides some data hiding that is useful for programming purposes but isn't much use for security. Another example of a security hole: >>> def f(): ... print "f()" ... >>> exec "from __main__ import f; f()" f() >>> Rob. -- http://www.victim-prime.dsl.pipex.com/ -- http://mail.python.org/mailman/listinfo/python-list
