[ruff] New rule unnecessary-if (RUF050)

[seroperson]

Summary

Implements the unnecessary_if rule (RUF050), which detects bare if statements (no elif/else) where the body contains only pass or ... and the condition is side-effect-free.

The rule complements RUF047 (needless-else): when all branches of an if/else are empty, RUF047 removes the empty else on the first fix pass, and RUF050 removes the remaining empty if on the next pass.

It also addresses #9472: when F401 removes unused imports from conditional blocks (e.g., if TYPE_CHECKING or if sys.version_info guards), the empty if blocks left behind are cleaned up by RUF050, and the now-unused guard imports are then removed by F401 on subsequent fix iterations.

⚠️ It's kind of superset of TC005, but handles not only TYPE_CHECKING guards. Both TC005 and RUF050 triggers on empty TYPE_CHECKING guard and I'm unsure how to be with it. Maybe skip TYPE_CHECKING guards in RUF050?

⚠️ It will "fix" the following the __bool__-like blocks: https://github.com/pandas-dev/pandas/blob/389051d090579550f1a1259855f486d7f7e2159e/pandas/tests/generic/test_generic.py#L147-L150
That's how contains_effect works and it seems a legit limitation, so I'm unsure how we should handle this (and should we at all). Should we make this fix unsafe because of this?

Conditions with side effects (function calls, await, yield, walrus operator) are skipped entirely, so the fix is always safe. If the body contains a comment (inline, own-line, or trailing), the rule also doesn't fire.

Test Plan

• RUF050.py - main rule test with error cases and non-error.
• unnecessary_if_and_needless_else - test function, which checks how RUF047 and RUF050 work together.
• unnecessary_if_and_unused_import - test function, which checks how F401 and RUF050 work together.

Closes #9472

Closes #13929 ❓

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

ℹ️ ecosystem check detected linter changes. (+10 -0 violations, +0 -0 fixes in 8 projects; 48 projects unchanged)

apache/airflow (+2 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview --select ALL

+ providers/amazon/tests/system/amazon/aws/example_bedrock_batch_inference.py:112:9: RUF050 [*] Empty `if` statement
+ shared/configuration/tests/configuration/test_parser.py:591:9: RUF050 [*] Empty `if` statement

apache/superset (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview --select ALL

+ superset/db_engine_specs/gsheets.py:518:13: RUF050 [*] Empty `if` statement

binary-husky/gpt_academic (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview

+ request_llms/embed_models/openai_embed.py:84:1: RUF050 [*] Empty `if` statement

langchain-ai/langchain (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview

+ libs/langchain/tests/unit_tests/test_imports.py:39:21: RUF050 [*] Empty `if` statement

pandas-dev/pandas (+2 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview

+ pandas/tests/generic/test_generic.py:148:13: RUF050 [*] Empty `if` statement
+ pandas/tests/indexes/test_any_index.py:19:9: RUF050 [*] Empty `if` statement

reflex-dev/reflex (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview

+ tests/units/vars/test_dep_tracking.py:279:13: RUF050 [*] Empty `if` statement

rotki/rotki (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview

+ rotkehlchen/icons.py:30:1: RUF050 [*] Empty `if` statement

zulip/zulip (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --no-fix --output-format concise --preview --select ALL

+ zerver/tests/test_validators.py:34:1: RUF050 [*] Empty `if` statement

Changes by rule (1 rules affected)

code	total	+ violation	- violation	+ fix	- fix
RUF050	10	10	0	0	0

[MichaReiser]

Thank you. This overall looks great. I've only a few smaller comments

[MichaReiser]

Can you add an example where the if is the only statement within a suite like what we see in https://github.com/pandas-dev/pandas/blob/389051d090579550f1a1259855f486d7f7e2159e/pandas/tests/generic/test_generic.py#L148

with pytest.raises(ValueError, match=msg):
    if obj1:
        pass

[seroperson]

I've added, but this intentionally kind of "wrongly" (specifically for this case) results in:

with pytest.raises(ValueError, match=msg):
    pass

Python's if statement calls obj1.__bool__() to evaluate the condition, and pandas overrides __bool__ on DataFrame to raise ValueError. That's what this check asserts (that you can't do if some_data_frame: checks in your code).

With RUF050 auto-fix this test will fail and I'm not sure what we can do about it (I've mentioned this in PR's description initially).

Shortly, in current implementation the situation is so:

• We strip if Var: pass entirely.
• Side-effects like if foo(): pass result in foo().
• But technically if Var: pass is actually if Var.__bool__(): pass, which has side-effect and logically must result in Var.__bool__(). But that's weird.

So we either should consider this limitation intended and expect users to set # noqa: RUF050 where they expect some side-effect from __bool__() (which is quite very edge case to be honest), or maybe make the whole rule unsafe.

[MichaReiser]

Let's mention this caveat in the documentation. This should be very uncommon and we can always change it if many users are running into this.

[MichaReiser]

Let me know when you updated the documentation so that we can merge this PR.

[seroperson]

@MichaReiser done

[MichaReiser]

⚠️ It's kind of superset of TC005, but handles not only TYPE_CHECKING guards. Both TC005 and RUF050 triggers on empty TYPE_CHECKING guard and I'm unsure how to be with it. Maybe skip TYPE_CHECKING guards in RUF050?

Skipping TYPE_CHECKING blocks for now seems reasonable. It gives people the option to keep empty TYPE_CHECKING blocks if they prefer so.

[MichaReiser]

Thank you