Changing runtime state

[orientalperil]

Why is it that when I assign variables in the shell, these assignments are not reflected in the running programme? When I use pdb, I am able to freely change the values of variables and execute the programme with my changes.

Is this ability omitted intentionally?

Is it possible to add?

[inducer]

pudb and pdb use literally the same code to implement user commands/variable assignments. It turns out that this is fine (and works in pudb and pdb) within the global scope, where variables are kept in a dict, but it very nearly can't be done inside a function, where variables are kept in the VM stack.

If you know otherwise, please let me know.

[orientalperil]

I see. It sounds like a hard problem.

[kgabor]

As of Python 3.6.6 this seems to be resolved in pdb, it can change local variables.

[inducer]

Interesting. In that case, could you look into what pdb does to make that happen? We should be able to just steal what they do.

[asmeurer]

I don't see any obvious changes to bdb.py or pdb.py related to this.

[inducer]

@kgabor I can't reproduce this. I use this script to test, on Python 3.7.3rc1.

def f():
    i = 0

    while i < 10:
        i = i+1
        print(i)

f()

I use pdb to step into f until I'm in the middle of the loop, then say

(Pdb) s i = 20
> /home/andreas/tmp/zzz.py(5)f()
-> i = i+1
(Pdb) p i
1

which seems to contradict your statement.

[disketten]

I think this should be re-opened as it works in pdb.

When I test the code from https://github.com/inducer/pudb/issues/26#issuecomment-486882575, I can modify i just fine:

$ python -m pdb zzz.py
> .../zzz.py(1)<module>()
-> def f():
(Pdb) n
> .../zzz.py(8)<module>()
-> f()
(Pdb) s
--Call--
> .../zzz.py(1)f()
-> def f():
(Pdb) n
> .../zzz.py(2)f()
-> i = 0
(Pdb) n
> .../zzz.py(4)f()
-> while i < 10:
(Pdb) n
> .../zzz.py(5)f()
-> i = i+1
(Pdb) n
-> print(i)
(Pdb) i
1
(Pdb) i = 20
(Pdb) i
20
(Pdb) n
20
> .../zzz.py(4)f()
-> while i < 10:

This is on Python 3.10.8.

[asmeurer]

It looks like pdb achieves this by making a copy of f_locals https://github.com/python/cpython/blob/aa8b58cb33826bd2b1a1de631ebcd6a5353eecb5/Lib/pdb.py#L299. Otherwise, frame.f_locals recomputes the locals every time it is accessed, wiping any changes that were made to it.

[asmeurer]

I have an in progress PR for a similar change for pudb to https://github.com/inducer/pudb/pull/571. It requires more testing, and also some cleaning up for the different shells.