The python code: 
```
class _tri(object):
    infts = '(+/-)inf, (+/-)infty, (+/-)infinity'

    strange_failing = {x+s.replace('(+/-)',''):None for x in ('+','-','') for s in infts.split(', ')}
```

gives a `global name 'infts' is not defined` exception, when normal dictionary comprehensions (without nested loops) and regular nested for-loops work perfectly well.

For a complete shell session and more illustrative example in versions 2.7.15 and 3.6.4 see: https://pastebin.ubuntu.com/p/9Pg8DThbsd/

I think this is explained in the below answers with example that the left most loop in the comprehension has access to the class variables which the nested comprehensions don't have

https://stackoverflow.com/a/22692274/2610955

https://stackoverflow.com/a/13913933/2610955

Thanks

This looks like a duplicate of issue3692.

But the simpler dictionary compprehension `{s.replace('(+/-)',''):None for s in infts.split(', ')}` works perfectly. Shouldn't that also give the error if it was a scope issue?

Okay, so we're a in another scope inside the dictionary comprehension (all comprehensions for that matter), and only one symbol is passed to the inside.

That's why `strange_reversed_working = {x+s.replace('(+/-)',''):None for x in infts.split(', ') for s in ('+','-','')}` functions, but if you reverse the order it does not. That's a real trap.

Updated example with reversed variable order for reference. This really seems to be related to issue3692, but really not the same thing.

IMHO both `a` and `b` should be passed in a situation like this:

````
a = range(5)
b = range(3)
c = [x+y for x in a for y in b]
````