single_range_in_vec_init stronger than expected, when type is known

[matzemathics]

Summary

The single_range_in_vec_init warning seems a bit overly cautious, if the type resulting from an invocation of a vec! macro is already known "from the outside", e.g. via a type annotation or when it is used as an argument / return value.

For example

let foo: Vec<Range<usize>> = vec![1..5]; // causes "warning: a `Vec` of `Range` that is only one element"

In this case clippy might infer, that the user indeed wanted a Vec<Range<...>>, because otherwise the program could not even have type checked. The same is true for this example:

fn barify(v: Vec<Range<usize>>) {
  ...
}

let v = vec![3..42]; // causes "warning: a `Vec` of `Range` that is only one element"
barify(v);

One explicit way to remedy this is using explicit lengths:

let foo = vec![1..5; 1]; // OK

although I am slightly worried that tools like cargo-fmt might at some point opt to remove that.

Lint Name

single_range_in_vec_init

Reproducer

I tried this code:

use std::ops::Range;

fn main() {
    let _foo: Vec<Range<usize>> = vec![1..5];
}

I saw this happen:

warning: a `Vec` of `Range` that is only one element
 --> src/main.rs:4:35
  |
4 |     let _foo: Vec<Range<usize>> = vec![1..5];
  |                                   ^^^^^^^^^^
  |
  = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#single_range_in_vec_init
  = note: `#[warn(clippy::single_range_in_vec_init)]` on by default
help: if you wanted a `Vec` that contains the entire range, try
  |
4 |     let _foo: Vec<Range<usize>> = (1..5).collect::<std::vec::Vec<usize>>();
  |                                   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
help: if you wanted a `Vec` of len 5, try
  |
4 |     let _foo: Vec<Range<usize>> = vec![1; 5];
  |                                        ~~~~

warning: `single_range` (bin "single_range") generated 1 warning

I expected to see this happen:
No warnings produced.

Version

No response

Additional Labels

No response

[Centri3]

Ignoring it for arguments when it's a local would likely be pretty tricky, though type annotations should definitely be checked

[ArthurCose]

Something not currently covered by the PR:

use std::ops::Range;

struct A {
    ranges: Vec<Range<usize>>,
}

struct B(Vec<Range<usize>>);

fn main() {
    A { ranges: vec![0..5] };
    B(vec![0..5]);
}

The correct type of a struct's field is also known "from the outside".

[Centri3]

Should be simple enough to fix

[Centri3]

It ignores ExprFields now. This also means something like Vec<&dyn Any> will be skipped as well but I think due to the rarity of both that and passing a Range<usize> to it, it should be fine. (It's also debatable whether that should even be linted.)

[djc]

I'm also running into this in Quinn, in the tests for the RangeSet::replace() method. This definitely seems like a false positive.

[avantgardnerio]

This is coming up with a lot of false positives for our project as well. I'm concerned that it is valid to have/want a Vec with a single Range, and that the suggestions to correct it would alter functionality. If anyone goes and applies them blindly, they might cause subtle bugs.

[Centri3]

The help message should probably be altered to make sure this doesn't happen, since right now it acts as though this is almost certainly what they want

[synek317]

It also causes quite annoying issues in my project. I completely don't understand the intention behind this lint and why is it enabled by default.

[bugadani]

This lint doesn't tell me what to do if I actually WANT an array/vec of ranges, which makes its strength even more bothersome.

[smalis-msft]

Agreed with everyone here, in our case it's only triggering on locals that are immediately passed to function calls that wouldn't typecheck otherwise. Should this lint be downgraded to a category that isn't warn-by-default?