Consider the following script:

    # import pdb; pdb.set_trace()
    # import re
    def f():
        import re
        print((lambda: re.findall('a', 'aaa'))())
        import pdb; pdb.set_trace()
        print('test')
    f()

When you run it and try to evaluate `(lambda: re.findall('a', 'aaa'))()`, you get:

    ['a', 'a', 'a']
    > /home/yuri/_/1.py(7)f()
    -> print('test')
    (Pdb) (lambda: re.findall('a', 'aaa'))()
    *** NameError: name 're' is not defined
    (Pdb) import re
    (Pdb) (lambda: re.findall('a', 'aaa'))()
    *** NameError: name 're' is not defined
    (Pdb) 

With the commented out breakpoint it works:

    > /home/yuri/_/a.py(3)<module>()
    -> def f():
    (Pdb) import re
    (Pdb) (lambda: re.findall('a', 'aaa'))()
    ['a', 'a', 'a']
    (Pdb) 

Also it works with uncommented global import and second breakpoint:

    ['a', 'a', 'a']
    > /srv/http/sl/makosh/a.py(7)f()
    -> print('test')
    (Pdb) (lambda: re.findall('a', 'aaa'))()
    ['a', 'a', 'a']
    (Pdb) 

From what I can see the issue occurs when there's no `re` in `globals` argument here: https://github.com/python/cpython/blob/v3.6.4/Lib/pdb.py#L376

I've run into it when trying to grep some object's attribute names, like:

    !list(filter(lambda x: re.search('class', x), dir(__name__)))

This can be traced back to the following issue:

>>> c = compile("(lambda: re.findall('a', 'aaa'))()", "<stdin>", "single")
>>> import re as rea
>>> exec(c, None, {'re': rea})
NameError: name 're' is not defined.

Seeing disassembly of the compiled code, it used LOAD_GLOBAL to get re. This seems to be the problem.

Sorry. I didn't finish my last message. Is the behaviour described in that message a bug? If yes, that would explain the original behaviour. If no, why not?