dead_code lint does not trigger for tuple structs with unread fields

[shepmaster]

This code:

fn main() {
    let _ = Example { u: 0 };
}

struct Example {
    u: u8,
}

Produces this warning:

warning: field is never read: `u`
 --> src/main.rs:6:5
  |
6 |     u: u8,
  |     ^^^^^
  |

However, switching to a tuple struct:

fn main() {
    let _ = Example(0);
}

struct Example(u8);

Does not report a warning.

Original report misleadingly focusing on Debug

fn main() {
    dbg!(Example { a: 0 });
}

#[derive(Debug)]
struct Example {
    a: u8,
}

produces

warning: field is never read: `a`
 --> src/main.rs:7:5
  |
7 |     a: u8,
  |     ^^^^^
  |
  = note: `#[warn(dead_code)]` on by default

However, switching to a tuple struct:

fn main() {
    dbg!(Example(0));
}

#[derive(Debug)]
struct Example(u8);

does not produce the warning.

/cc @FabianWolff
/cc #85200; #84647; #88900

Meta

Rustc 1.57.0 and 1.60.0-nightly (2022-01-10 89b9f7b)

[FabianWolff]

This has nothing to do with the #[derive(Debug)]:

fn main() {
    let _ = Example { u: 0 };
}

struct Example {
    u: u8,
}

warning: field is never read: `u`
 --> src/main.rs:6:5
  |
6 |     u: u8,
  |     ^^^^^
  |

but

fn main() {
    let _ = Example(0);
}

struct Example(u8);

gives no warning.

[shepmaster]

@FabianWolff oh dear. Thanks! I've updated the title and original post; sorry for the false ping.

[camelid]

Not a regression AFAIK, but seems like worth prioritizing.

[shepmaster]

Not a regression AFAIK

I agree. I did a quick test across Rust 1.{0,10,20,30,40,50}.0 and none of them reported the lint.

[camelid]

Thanks for checking!

[FabianWolff]

This behavior does indeed date back to the very beginning of the detection of dead struct fields (commit 0271224 in 2014). The culprit is line 621 in dead.rs:

rust/compiler/rustc_passes/src/dead.rs

Lines 618 to 625 in 72e74d7

	fn should_warn_about_field(&mut self, field: &hir::FieldDef<'_>) -> bool {
	let def_id = self.tcx.hir().local_def_id(field.hir_id);
	let field_type = self.tcx.type_of(def_id);
	!field.is_positional()
	&& !self.symbol_is_live(def_id)
	&& !field_type.is_phantom_data()
	&& !has_allow_dead_code_or_lang_attr(self.tcx, field.hir_id)
	}

Commenting out this line with the above tuple struct example gives:

warning: field is never read: `0`
 --> s2.rs:6:16
  |
6 | struct Example(u8);
  |                ^^
  |

I can open a PR for this, or is there a reason anyone here can think of why dead tuple struct fields shouldn't get a warning?

@rustbot claim

[camelid]

I can't think of one, so opening a PR seems good.

[danielhenrymantilla]

FWIW, I find the example with dbg! to be surprising: dbg! is debug-printing a struct, and thus reading its fields

[shepmaster]

@danielhenrymantilla see #85200; #84647; #88900