pdb loses local variable change after command longlist

[gaogaotiantian]

Bug report

In pdb, ll will clear the local variable change.

def main():
    a = 1
    breakpoint()
    print(a)

main()

-> print(a)
(Pdb) p a
1
(Pdb) !a = 2
(Pdb) p a
2
(Pdb) ll
  1     def main():
  2         a = 1
  3         breakpoint()
  4  ->     print(a)
(Pdb) p a
1
(Pdb) s
1
--Return--

As you can tell, a was changed through !a = 2 but it was reverted after ll. print(a) also prints the unmodified value. Without ll, everything works fine.

The reason lies in getsourcelines in pdb.py. In that function, it tried to access obj.f_locals, which will refresh the dict with PyFrame_FastToLocalsWithError as it's a property now.

if inspect.isframe(obj) and obj.f_globals is obj.f_locals:

As a result, the local variable changes will be erased.

The original reason to check if obj.f_globals is obj.f_locals is to check whether obj is an module frame. Now that it has side effects to pdb, we can do the check with:

if inspect.isframe(obj) and obj.f_code.co_name == "<module>":

It might not be the most delicate way, but I can't think of a situation where this won't work.

I did this change locally and I have confirmed:

1. ./python -m test -j3 passed
2. The original bug is fixed
3. ll still prints the full file for module frame

I'll make a PR soon and please let me know if there are any concerns about the fix.

Your environment

• CPython versions tested on: 3.11.1
• Operating system and architecture: Ubuntu 20.04

Linked PRs

• GH-101673: Fix pdb bug that local variable changes lost after longlist #101674
• [3.11] GH-101673: Fix pdb bug where local variable changes are lost after longlist (GH-101674) #102632
• [3.10] GH-101673: Fix pdb bug where local variable changes are lost after longlist (#101674) #102633

[gaogaotiantian]

Hi all, it's been about a month since this issue is created. I think this issue is pretty straightforward and easy to reproduce. The fix in #101674 is easy to understand and well contained. Could someone give this issue/PR a review? Thanks.

[iritkatriel]

Any bugfix PR should include a unit test to prevent regressions.

[gaogaotiantian]

Hi @iritkatriel , thank you for the reminder! I've added the regression test for this bug - I confirmed that the test fails without the fix and passes with the fix.