clippy::needless_collect suggests rewriting to code that is semantically different

Summary

The needless_collect lint applied to VecDeque::drain diagnosed a call to collect, when the resulting vector was used in a subsequent iteration statement. However, it doesn't consider an intervening pop_back on the same VecDeque, which would alter the semantics of the program (removing the value from pop_back first, then draining the range, rather than draining the range, then using pop_back).

Lint Name

needless_collect

Reproducer

I tried this code:

#![deny(clippy::all)]

use std::collections::VecDeque;

fn test(vstack: &mut VecDeque<i32>, params: usize){
    let vstack_len = vstack.len();
    let pvals = vstack
        .drain((vstack_len - params)..)
        .collect::<Vec<_>>();
    let function = vstack.pop_back().unwrap();
    for (i, val) in pvals.into_iter().enumerate() {
        match i {
            0 => {}
            _ => panic!(),
        }
    }
    
}

I saw this happen:

  |
9  |         .collect::<Vec<_>>();
   |          ^^^^^^^
10 |     let function = vstack.pop_back().unwrap();
11 |     for (i, val) in pvals.into_iter().enumerate() {
   |                     ----------------- the iterator could be used here instead
   |
note: the lint level is defined here
  --> src/lib.rs:1:9
   |
1  | #![deny(clippy::all)]
   |         ^^^^^^^^^^^
   = note: `#[deny(clippy::needless_collect)]` implied by `#[deny(clippy::all)]`
   = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#needless_collect
help: use the original Iterator instead of collecting it and then producing a new one
   |
7  ~     
8  |     let function = vstack.pop_back().unwrap();
9  ~     for (i, val) in vstack
10 ~         .drain((vstack_len - params)..).enumerate() {
   |

I expected to see this happen:
No Lint

Version

rustc 1.57.0 (f1edd0429 2021-11-29)
binary: rustc
commit-hash: f1edd0429582dd29cccacaf50fd134b05593bd9c
commit-date: 2021-11-29
host: x86_64-unknown-linux-gnu
release: 1.57.0

Additional Labels

No response

Start by locating the needless_collect lint implementation and its existing tests. Add the provided VecDeque::drain and pop_back reproducer as a regression case, then verify the lint does not suggest replacing the collection with a direct iterator when that would change evaluation order.