Why are closures broken within exec?
In Python 2.6,
>>> exec "print (lambda: a)()" in dict(a=2), {}
2
>>> exec "print (lambda: a)()" in globals(), {'a': 2}
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<string>", line 1, in <module>
File "<string&g开发者_Python百科t;", line 1, in <lambda>
NameError: global name 'a' is not defined
>>> exec "print (lambda: a).__closure__" in globals(), {'a': 2}
None
I expected it to print 2
twice, and then print a tuple with a single cell
. It is the same situation in 3.1. What's going on?
When you pass a string to exec
or eval
, it compiles that string to a code object before considering globals or locals. So when you say:
eval('lambda: a', ...)
it means:
eval(compile('lambda: a', '<stdin>', 'eval'), ...)
There's no way for compile
to know that a
is a freevar, so it compiles it to a global reference:
>>> c= compile('lambda: a', '<stdin>', 'eval')
>>> c.co_consts[0]
<code object <lambda> at 0x7f36577330a8, file "<stdin>", line 1>
>>> dis.dis(c.co_consts[0])
1 0 LOAD_GLOBAL 0 (a)
3 RETURN_VALUE
Therefore to make it work you have to put a
in the globals and not the locals.
Yeah, it's a bit dodgy. But then that's exec
and eval
for you I suppose... they're not supposed to be nice.
精彩评论