Repository navigation
inlined comprehension and closed over variables have buggy interactions #156664
Description
Activity
CC @carljm
Surprised this has never been caught before! I even ran a fuzzer when we were implementing PEP 709 to find things like this. I'm attempting a fix now.
An analogous case that is also broken in 3.12+:
def test_inlined_comprehension_then_nested_function(): x = 3 def f(): [x for x in [1]] def g(): return x return g() assert f() == 3 test_inlined_comprehension_then_nested_function()
And a more complicated version:
def test_captured_target_then_sibling_comprehension(): x = 3 def f(): funcs = [lambda: x for x in [1]] return funcs[0](), [x for _ in [1]] assert f() == (1, [3]) test_captured_target_then_sibling_comprehension()
Bug report
Bug description:
We are a bit late to report this, but the following of PyPy's unit tests got broken by inlined comprehensions in CPython 3.12:
this passes fine in CPython 3.11 but breaks like this in 3.12 onwards, including on main:
$ ./python /tmp/bug.py Traceback (most recent call last): File "/tmp/bug.py", line 10, in <module> test_inlined_comprehension_private_name_reuse() ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^ File "/tmp/bug.py", line 8, in test_inlined_comprehension_private_name_reuse assert f() == [3] ~^^ File "/tmp/bug.py", line 6, in f return [x for _ in [1]] ^ UnboundLocalError: cannot access local variable 'x' where it is not associated with a valueI wonder whether that is the kind of bug that would be easier to fix when doing #124697
CPython versions tested on:
CPython main branch
Operating systems tested on:
No response
Linked PRs