FP: `used_underscore_items` lints on debatably legitimate use-cases

Summary

I feel the new used_underscore_items lint is not universally applicable and would best live in restriction rather than pedantic, as there are sometimes legitimate use-cases for prefixing item names with a _.

Lint Name

used_underscore_items

Reproducer

As motivation, I'll give two examples where prefixing method names with _ makes sense:

First, as a way to reduce the impact of monomorphization on binary size for functions that have large bodies:

pub fn my_method(&self, param: impl AsRef<str>) {
    self._my_method(param.as_ref())
}

fn _my_method(&self, param: &str) {
    // large function body here
}

Second, as a way to introduce non-breaking changes when adding a new argument to a function. If we have the following initial code:

pub fn my_method(&self, param1: u32) {
    // impl
}

...and we want to add support for a second parameter without making breaking changes, we could transform it in the following way:

pub fn my_method(&self, param1: u32) {
    self._my_method(param1, None)
}

pub fn my_method_2(&self, param1: u32, param2: u64) {
    self._my_method(param1, Some(param2))
}

fn _my_method(&self, param1: u32, param2: Option<u64>) {
    // impl
}

In both cases I feel that naming the private method with a prefixed underscore is a valid choice to make, in order to preserve the name and to not conflict with the public wrapper method. Granted, maybe this still means it should be a pedantic lint, but either way I feel these two cases qualify as false positives.

Version

rustc 1.83.0-nightly (ed04567ba 2024-09-28)
binary: rustc
commit-hash: ed04567ba1d5956d1080fb8121caa005ce059e12
commit-date: 2024-09-28
host: x86_64-unknown-linux-gnu
release: 1.83.0-nightly
LLVM version: 19.1.0

Additional Labels

No response

Start with the used_underscore_items lint entry point and its existing tests, then compare the two reproducer patterns in the issue with the lint’s current behavior. Review the discussion about pedantic versus restriction classification and determine the agreed scope. Done means the accepted behavior and lint classification are reflected consistently in the implementation and tests.