Wrong tuple lengths being documented.

[cbrownsey]

variadics_please version

Version 1.1.0.

What you did

In a crate I'm working on I saw that the implementation was for up to 17 items. This drew my eye because the 16 length limit is a limitation of the C library I'm using. A reproduction of the issue is below and can be run with cargo +nightly -Zscript ./repro.rs.

---
[dependencies]
variadics_please = "1.1.0"
---

macro_rules! fake_impl {
    ($(#[$meta:meta])* $($ty:ident),*) => {
        let s = stringify!($($meta)*);
        if !s.is_empty() {
            println!("{s}");
        }
    }
}

fn main() {
    variadics_please::all_tuples!(
        #[doc(fake_variadic)]
        fake_impl,
        0,
        1,
        T
    );
}

What went wrong

Despite only running on the tuples () and (T,), this snipped outputs the following:

cfg_attr(any(docsrs, docsrs_dep), doc(fake_variadic))
cfg_attr(any(docsrs, docsrs_dep), doc =
"This trait is implemented for tuples up to 2 items long.")

This is incorrect, as it will only implement the trait for items up to 1 item long.

Additional information

With a brief grep in the bevy repository, this issue looks like it affects at least twenty different trait implementations.

The source of this issue looks to be a double counting in each of the macro functions in the line let len = 1 + input.end - input.start;. Although this is used as an endpoint in a loop, the result of that iteration is ignored as the next iteration is over input.start..=input.end, which is why calling all_tuples with a range of 0 to 1 hasn't been outputting three implementations.