[possibly-undefined] False-positive with nested try block and finally

@ilinum

Bug Report
Small false-positive I noticed after #14114.
Seems to be a scoping issue with finally.

To Reproduce

try:
    try:
        msg = some_func()
    except Exception:
        raise

    print(msg)  # Name "msg" may be undefined  [possibly-undefined]

finally:
    pass

print(msg)  # <-- should probably be emitted here
# EDIT: exception is re-raised - no error should be emitted

The first print will always safely execute since the inner Try block raises. If it throws an uncaught exception, the print will be skipped to the finally block and only the second print will execute which would then raise an error itself as msg would be undefined.

Your Environment

Mypy version used: mypy 1.0.0+dev.96ac8b3e71c743b02ea4e3c84da0248659e40f82 (compiled: no)
Mypy command-line flags: --enable-error-code possibly-undefined
Python version used: 3.11

If some_func raises an exception, the second print will never be executed because the exception will be re-raised at the end of the finally block. So I think that means no error should be generated for the final line of the code sample.

Yep, good find! This only occurs with nested statements because we have to do a ~~hack~~ clever trick with disable_branch_skip.

We can probably do something smarter here and only report errors in finally blocks when we're in disable_branch_skip mode.

I don't really have time to work on this right now, so happy for someone else to pick it up. Otherwise, I could probably start in February.

Not entirely sure if this is the same issue or a separate issue, but we are also seeing this where an except then breaks from a loop:
https://github.com/aio-libs/aiohttp/blob/63f83a0f916ee5030ee0f2dad3ba999b325c167c/aiohttp/web_protocol.py#L540-L549

Is there still an issue here? If so, could someone provide a self-contained minimal code sample?

The original example still reproduces (again, remember to enable the error code). It currently produces possibly-undefined on line 7, but not on line 12. As you mentioned above, there should be no errors, as the exception is reraised.

In the linked example from aiohttp, we do a break in the exception handler, so also should not get any errors. However, a minimal example with break doesn't produce an error, so not sure why it behaves differently on our code base.

Here’s the example as a playground link (using inline configuration):

https://mypy-play.net/?mypy=latest&python=3.11&gist=c74b25457397d60fe5f063151ceb5fee

Ah, the presence of the finally affects the output. Here's the aiohttp version with a break:

while True:
    try:
        try:
            msg = some_func()
        except Exception:
            break
        print(msg)  # Name "msg" may be undefined  [possibly-undefined]
    except Exception:
        pass
    finally:
        pass

here's a much simpler example that appears to be an exact duplicate:

def f(x: int) -> None:
    try:
        if x >= 5:
            y = 2
        else:
            raise AssertionError('unreachable')
        print(y)
    finally:
        print('finally!')

$ mypy t.py --enable-error-code=possibly-undefined
t.py:7: error: Name "y" may be undefined  [possibly-undefined]
Found 1 error in 1 file (checked 1 source file)

Another instance from python/typeshed#12151 (when using --platform win32):

import sys

try:
    if sys.platform == "win32":
        pip_exe = ""
        reveal_type(pip_exe)  # Revealed type is "builtins.str"
    else:
        pip_exe = ""
    reveal_type(pip_exe)  # Revealed type is "builtins.str"
                          # Name "pip_exe" may be undefined  [possibly-undefined]
finally:
    pass

Fix possibly-undefined false-positive with nesting

Fix possibly-undefined false-positive with nesting (#20276)

Navigation Menu

Repository navigation

[possibly-undefined] False-positive with nested try block and finally #14309

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

[possibly-undefined] False-positive with nested try block and finally #14309

Description

Activity

erictraut commented on Dec 17, 2022

ilinum commented on Dec 19, 2022

Dreamsorcerer commented on Jul 27, 2023

erictraut commented on Aug 14, 2023

Dreamsorcerer commented on Aug 14, 2023

andersk commented on Aug 14, 2023

Dreamsorcerer commented on Aug 14, 2023

asottile-sentry commented on Apr 3, 2024

srittau commented on Jul 12, 2024

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions