False positive consider-using-with in if/else block

[DudeNr33]

Bug description

Given a file a.py:

# pylint: disable=missing-docstring, invalid-name
a = True
if a:
    file_handle = open("foo")
else:
    file_handle = open("bar")
with file_handle:
    print(file_handle.read())

(Configuration)

No response

Command used

Result of pylint a.py:

************* Module a
a.py:4:18: R1732: Consider using 'with' for resource-allocating operations (consider-using-with)

------------------------------------------------------------------
Your code has been rated at 8.33/10 (previous run: 8.33/10, +0.00)

(Expected behavior)

No message consider-using-with, as file_handle is used in a subsequent with block.
#4721 implemented changes that already suppress this message for things like:

file_handle = open("foo")
with file_handle:
    print(file_handle.read())

So naturally the example code above should also not trigger the message.

(Version affected)

pylint 2.9.5
astroid 2.6.5
Python 3.9.6 (default, Jun 28 2021, 19:24:41) 
[Clang 12.0.5 (clang-1205.0.22.9)]

(OS / Environment)

macOS BigSur 11.4

(Additional dependencies)

No response

[timmartin]

The issue seems to be that when the second assignment is encountered:

a = True
if a:
    file_handle = open("foo")
else:
    file_handle = open("bar")  # <--here

the checker thinks it's being reassigned, despite the fact that the program can't possibly hit both assignments in the same run. I've had a quick look at the existing checkers and I'm not sure if this is something that can be addressed easily. It doesn't seem like Pylint / Astroid as it stands has any logic for dealing with the case that something is conditionally defined in one arm of an if block.

For example:

if random.random() < 0.5:
    b = 6
else:
    pass

print(b)

doesn't report (possibly) uninitialised variable. Is this correct? If so, then this isn't really specific to the consider-using-with checker but will be a much wider change that would need to be made.

[DudeNr33]

You are definitely correct that the Problem is in the else clause. I have an idea how to solve it for this check, but that would not solve the general problem you described.

[DudeNr33]

The problem with the consider-using-with check is, that it intends to check if a variable containing a context manager is redefined before it is used in a with block, as you already mentioned.

Now for consider-using-with I would extend the logic of the check to see if the assignment is made inside an else block and if so double check if the previous assignment happened in the corresponding if block.

But that is just a solution to this check and not for others.
I tried to see how pylint would handle a normal case of a variable that is redefined before it is used (as e.g. PyCharm does it), but apparently this is not checked by pylint at all:

a = 1
a = 2  # <-- no message here
print(a)

Edit: might benefit from results of #4795.