Execution fails with NameError when trying to dynamically compile a Python program that uses a function in a comprehension, and the code is compiled inside a function call, rather than as a global statement.

Using the attached files to reproduce (in reproduce.zip):

python3 compile_in_function.py comprehension.py #fails
python3 compile_global.py comprehension.py # works
python2 compile_in_function.py comprehension.py #works with Python 2.7.12

Could you please post the examples instead of a zip file?  Zip files are hard to read on a mobile browser :)

Woops, sorry about that, makes sense. Below an example (same idea as the files):

```
Python 3.6.2 (default, Sep  9 2017, 13:27:06) 
[GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.38)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> src = """
... def f(x): return x
... [f(v) for v in [1,2,3]]
... """
>>> def comp():
...     exec(compile(src, filename='<str>', mode='exec'))
... 
>>> comp()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 2, in comp
  File "<str>", line 3, in <module>
  File "<str>", line 3, in <listcomp>
NameError: name 'f' is not defined
>>> exec(compile(src, filename='<str>', mode='exec'))
>>> 
```

List comprehension is irrelevant here, any nested function has the same problem. And the inner name doesn't have to refer to a function. Also, compilation as a separate step is not necessary.

So the minimal example is

    src = '''\
    a = 5
    def g(x): return a + x
    print(g(3))
    '''
    def comp(): exec(src)
    comp()

But I'm not sure if it is a bug at all. If you write explicitly

    exec(src, {})

you get a result 8 without problem. The docs say

    ...if the optional parts are omitted, the code is executed in the current scope.

And the current scope inside comp is "fast locals", where you can't really make new names. Maybe exec should raise an exception then, but it surely cannot just "work" inside fast locals scope.

This is indeed not a bug, it it a consequence of the scoping rules.

What surprises me is that it works without passing in an explicit scope in in 3.7.  I'm not sure what we changed that makes that true.

Hm... then it _is_ a bug (in 3.7). Or we have "fast locals and slow locals" nightmare again... :-/

I don't have 3.7 to test... what do you mean is working? jpc's example or mine?

Hmm.  I must have made a mistake when I ran (jpc's) test on 3.7.  It is failing with the NameError for me when I try it again.