[youdie006]

VecMap::insert_at documents a panic only for index > len. With an existing key and index == len, it removes the old entry first and then inserts into a vector that is one shorter, so it panics and the entry is gone:

let mut map = VecMap::from([("a", 1), ("b", 2)]);
map.insert_at(2, "a", 3);
// before: panics with "insertion index (is 2) should be <= len (is 1)";
//         if the panic is caught, the map is [("b", 2)]
// after:  Some((0, 1)), and the map is [("b", 2), ("a", 3)]

The fix checks index > len before anything is removed, and moves the entry to the end when index == len. The doc now says so. New keys and index > len behave as before. Across maps of length 1 to 4, with every existing key and every index up to len, only the 10 cases with index == len change.

If you'd rather keep the panic, like indexmap's shift_insert, which rejects index == len for an existing key, the alternative is to assert index < len before removing and adjust the doc. That would still fix the lost entry.

These pass: cargo test --workspace --all-features, cargo fmt --check, and cargo doc with -D warnings. I didn't run miri, but the change has no unsafe code.

Written with AI assistance (Claude); I have reviewed the change.

[martinohmann]

Thanks for fixing this!