Consider the following code:

#[derive(Copy, Clone)]
struct Foo {
    a: usize,
    b: usize,
    c: usize,
}

fn take_foo(foo: &Foo) {}

When targeting 64-bit Foo is 24 bytes and so doesn't trigger trivially_copy_pass_by_ref for take_foo.

However when targeting 32-bit Foo is 12 bytes, which is below our threshold of 16 bytes, and triggers the lint. Satisfying the lint doesn't make sense in the larger picture though, because Foo is still 24 bytes in the 64-bit world.

I ran into this issue in practice when adding our lint set to SimpleCSS.

Unfortunately our custom trivial-copy-size-limit can't be a multiplier of target_pointer_width even though the default value does that. So the solution is to make the lint apply only when targeting 64-bit. Can't do that in Cargo.toml so lib.rs it is. Luckily we don't care as much about this lint in the examples so it works out fine.