Fix let_unit_value with for loop iterating over units

[koivunej]

This will fix #1964, which simplifies to the code below getting lint let_unit_value:

for _ in &[()] {
  // anything
}

Fix is to modify is_from_for_desugar to detect HIR-level let _ = for _ in ... {} in addition to the previously detected HIR-level let _ = for something in ... {}. Also adds comments to explain the previous desugaring detection.

Does not add the test case suggested by @oli-obk:

Also add an example that would have triggered if the first if_let_chain were removed, so we don't have to hope dogfood will catch it but have an explicit test in the suite.

I did not add a new case because commenting out the first check (and disabling tests/dogfood.rs) will cause a large number of tests/ui/for_loop.rs to receive for_let_unit lints such as:

error: this let-binding has unit value. Consider omitting `let for _v in hm.iter() { } =`
   --> $DIR/for_loop.rs:234:5
    |
234 |     for _v in hm.iter() { }
    |     ^^^^^^^^^^^^^^^^^^^^^^^

Only ui-test changes are included as I didn't find any test cases for the is_from_for_desugar. I'd imagine something like the example below would be awesome but might not be possible at the moment (I don't know the rustc at all).

assert!(is_from_for_desugar(hir! { for _ in 0.. {} });

[koivunej]

This detects a variable binding ...

Should had been something like "This detects a _ binding ..." or "This detects an underscore binding". Latter might be more searchable or greppable?

[oli-obk]

This is fine. I think it would happen for others, too, like if you had for Foo { a } in bar {}

[koivunej]

Cannot see it matching this for-loop. is_from_for_desugar returns false for the following:

#![feature(plugin)]
#![plugin(clippy)]

fn main() {
    let b = vec![Foo { a: 42 }];

    for Foo { a } in b {
        println!("a: {}", a);
    }
}

struct Foo {
    a: u32
}

The `{:#?}` output of decl in above example

Spanned {
    node: DeclLocal(
        Local {
            pat: pat(8: b),
            ty: None,
            init: Some(
                expr(26: <[_]>::into_vec(box [Foo{a: 42,}]))
            ),
            id: NodeId(
                7
            ),
            hir_id: HirId {
                owner: DefIndex(
                    3
                ),
                local_id: ItemLocalId(
                    1
                )
            },
            span: ./example.rs:5:5: 5:32,
            attrs: ThinVec(
                None
            ),
            source: Normal
        }
    ),
    span: ./example.rs:5:5: 5:33
}

[koivunej]

For the previous comment I had the println!'s wrong. There are a couple of calls to is_from_for_desugar but the example for Foo { a } in bar is handled by the original check in is_from_for_desugar.

[oli-obk]

I think that's a different DeclLocal than the one we're looking for, but it should probably also contain a source other than Normal. I'll see if I can produce a PR for rustc.

I think it'll call is_from_for_desugar if you have a let a = 42; outside the for loop, which will trigger the shadowing lints.

This PR is fine as it is, since this is a non-issue for the let_unit_value lint anyway

[oli-obk]

Thanks! This is some great detective work you did there