fix(useless_format): fire on wrapped in a block-producing macro

[shulaoda]

Summary

Fixes a false negative in useless_format: the lint silently failed to fire on format!("{}", s) when the call is the tail expression of a block produced by another macro. This affects rustc's entire define_helper!-generated family (with_forced_trimmed_paths!, with_no_trimmed_paths!, with_no_queries!, etc).

Spotted during review of #17058:

Even as a macro argument, using format!("{}", some_string) should trigger it.

Reproducer

#![feature(decl_macro)]
#![warn(clippy::useless_format)]

macro_rules! plain_mr { ($e:expr) => { $e }; }
macro_rules! block_mr { ($e:expr) => { { let _g: i32 = 0; $e } }; }
pub macro plain_dm($e:expr) { $e }
pub macro block_dm($e:expr) { { let _g: i32 = 0; $e } }

fn s() -> String { String::from("x") }

fn main() {
    let _ = format!("{}", s());                  // (1) bare         => lints
    let _ = plain_mr!(format!("{}", s()));       // (2) mr, no block => lints
    let _ = block_mr!(format!("{}", s()));       // (3) mr + block   => MISS (bug)
    let _ = plain_dm!(format!("{}", s()));       // (4) dm, no block => lints
    let _ = block_dm!(format!("{}", s()));       // (5) dm + block   => MISS (bug)
}

Before this PR, cases (3) and (5) are silently missed. After this PR all five fire as expected. Note that the discriminator is block-wrapping, not macro_rules vs decl_macro.

Root cause

clippy_lints/src/format.rs previously used root_macro_call_first_node, which requires first_node_in_macro == Some(ExpnId::root()). For block_dm!(format!(...)), the format! HIR parent is the Block emitted by block_dm, whose span lives in block_dm's expansion (sibling to format!'s). first_node_in_macro returns Some(block_dm.expn) rather than Some(root), and the lint bails.

Fix

Replace the gate with:

if let Some(macro_call) = matching_root_macro_call(cx, expr.span, sym::format_macro)
    && first_node_in_macro(cx, expr).is_some_and(|p_expn| p_expn != macro_call.expn)

• matching_root_macro_call preserves hygiene. The outermost macro on expr.span's backtrace must be format!, so a format! written inside another macro's body (where the rewrite would target the macro definition) is still ignored.
• first_node_in_macro(..).is_some_and(|p| p != macro_call.expn) preserves single-firing. expr must be the outermost node of format!'s expansion. Without p != macro_call.expn, deeper nodes whose parent is also in format!'s expansion (including internal format_args! invocations) would slip through and the lint would fire multiple times per call.

Test changes

• tests/ui/format.{rs,fixed,stderr}: added #![feature(decl_macro)] and a block_wrap module covering all four pass-through and block-wrap combinations across macro_rules! and decl_macro, as regression coverage.
• tests/ui/unused_format_specs.{rs,1.fixed,2.fixed}: added clippy::useless_format to the allow-list. The relaxed gate also starts firing on println!("{:.3}", format!("abcde"))-style cases, which appear in this test's .fixed outputs (after unused_format_specs suggests format_args! to format!). This is a latent true positive previously masked by the strict gate. The test is scoped to unused_format_specs and should not be entangled with another lint's coverage. The same pattern is already used by tests/ui/format.rs itself, which allows other unrelated lints.

Verification

• cargo test --test compile-test: 1764 UI tests + 188 fixed checks + 46 ui-cargo tests pass. 0 duplicate diagnostics.
• cargo test --test dogfood: catches the same false negative in clippy's own unnecessary_literal_unwrap.rs (the line that useless_format misses format! inside block-wrapping macros #17059 is fixing manually), demonstrating the fix on real-world code.

Related

• Motivating discussion: Remove redundant format!("{}", ..) around def_path_str in unnecessary_literal_unwrap #17058.
• Companion manual fix that this PR's lint now catches automatically: useless_format misses format! inside block-wrapping macros #17059.
• Test file: tests/ui/format.rs, see the new mod block_wrap at the bottom.

changelog: [useless_format] no longer misses format! calls that are the tail expression of a block produced by another macro (e.g. rustc's with_forced_trimmed_paths!).

[rustbot]

r? @Jarcho

rustbot has assigned @Jarcho.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: 7 candidates
• 7 candidates expanded to 7 candidates
• Random selection from Jarcho, llogiq, samueltardieu

[shulaoda]

r? @samueltardieu

[samueltardieu]

Can you include a commit before the current one which removes the use of format!() in unnecessary_literal_unwrap.rs? This way, the build stays clean at any point.

View changes since this review

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[shulaoda]

@rustbot ready

[shulaoda]

@samueltardieu Thanks for the patient review 🙏