//! Each misuse of an attribute, refused at the attribute.
//!
//! The expected output next to each case is the derive's own message, spanned
//! where the user wrote the thing it refuses. These are stable across
//! compiler versions because none of them reaches the macros or the type
//! checker: the derive refuses before it expands. That placement is the thing
//! being tested, so the comparison stays `nocompile`'s default `Exact`, which
//! keeps the span art a `Brief` run would drop.
//!
//! Skipped on Windows, where `nocompile` declines to claim support: it folds
//! separators and strips `\r` under unit test, but its author has no Windows
//! machine to confirm that on. Untested rather than known broken, so the test
//! is `ignore`d there rather than compiled out -- `cargo test -- --ignored` on
//! a Windows box is what would settle it. The other two CI platforms run the
//! suite, and nothing about a rejected attribute is host-specific.