fix: never remove parens from prefix ops with valueless return/break/continue

[Aditya-PS-05]

fixes #21063

The Remove redundant parentheses was suggesting removal of parentheses in expressions like !(return), !(break), and !(continue), which would result in invalid syntax.

So, added a case in Expr::needs_parens_in() which checks for two things:
1.) Prefix operator
2.) is_ret_like_with_no_value() which handles return, yield, break, continue, become

If both are present, the parenthesis are not removed.

Added test also.

[asuto15]

I opened PR #21090 for the same issue, but your approach looks smarter and the test coverage is broader than mine.

To the maintainers: if this solution is preferred, I can close my PR (#21090). Please let me know your preference.

[Aditya-PS-05]

@asuto15 actually I started working on the issue and completed earlier than you but could not open a pr due to time constraint. So, I opened it anyway.

[ShoyuVanilla]

Yeah, this one is a bit cleaner than #21090 (though I think both are great), so I'm going to merge this one as @asuto15 suggested.
But in general, I think the person assigned to the issue should have priority.

[A4-Tacks]

This does not solve issue, such as this test:

check_assist_not_applicable(
    remove_parentheses,
    r#"fn main() { let _x = true && $0(return) || true; }"#,
);
check_assist(
    remove_parentheses,
    r#"fn f() { cond && !$0(return) }"#,
    r#"fn f() { cond && !return }"#,
);

[A4-Tacks]

If that’s the intended behavior

Yes, this is intended behavior. I missed typing a character

[asuto15]

Thank you!

One more question for the maintainers @ShoyuVanilla and @A4-Tacks:

Would you prefer that we treat all valueless control-flow expressions (return, break, continue) uniformly?

For example, allowing:

fn f() { !($0(return)) }   // -> !return
fn f() { !($0(break)) }    // -> !break
fn f() { !($0(continue)) } // -> !continue

(except the return || case, which must remain blocked)

This would give a simple, consistent rule.
If this direction is acceptable, I can adjust the PR(#21090) accordingly.

[A4-Tacks]

valueless control-flow expressions

Consider this test:

check_assist_not_applicable(
    remove_parentheses,
    r#"fn main() { let _x = true && $0(return false) || true; }"#,
);
check_assist(
    remove_parentheses,
    r#"fn f() { cond && !$0(return false) }"#,
    r#"fn f() { cond && !return false }"#,
);

[ShoyuVanilla]

This does not solve issue, such as this test:

check_assist_not_applicable(
    remove_parentheses,
    r#"fn main() { let _x = true && $0(return) || true; }"#,
);
check_assist(
    remove_parentheses,
    r#"fn f() { cond && !$0(return) }"#,
    r#"fn f() { cond && !return }"#,
);

Oh, good catch

[Aditya-PS-05]

Sorry I missed this discussion. Does this pr not fixes the issue or any edge case?

[A4-Tacks]

Does this pr not fixes the issue or any edge case?

The main issue is BinExpr instead of PrefixExpr