[BenjaminLucier]

Closes #1122.

Problem

Wrapping the arguments of s![] in brackets is an easy mistake, and the error doesn't point at it:

error[E0277]: the trait bound `[std::ops::Range<{integer}>; 1]: ndarray::SliceNextDim` is not satisfied

It names a hidden trait and never mentions s![] or what the macro accepts.

Fix

Add #[diagnostic::on_unimplemented] to SliceNextDim, the trait that s![] requires of each argument. Now s![[1..3, 0..2]] reports:

error[E0277]: `[std::ops::Range<{integer}>; 2]` cannot be used as an argument to the `s![]` macro
  |
  |     let _ = a.slice(s![[1..3, 0..2]]);
  |                     ^^^^^^^^^^^^^^^^ not an index, range, `Slice`, or `NewAxis`
  |
  = note: each comma-separated argument to `s![]` must be an index (`isize`, `usize` or `i32`), a range such as `1..3` or `..`, a `Slice`, or `NewAxis`, optionally followed by `;step`
  = note: do not wrap the arguments in brackets or parentheses: write `s![1..3, ..]`, not `s![[1..3, ..]]`

Tuples such as s![(1..3, 2)] get the same message. #[diagnostic::on_unimplemented] has been stable since Rust 1.78, below the MSRV of 1.87.

The s! docs now also explain the mistake, with a compile_fail example, as the issue suggested.

Two related mistakes are not covered, because they fail earlier with a different error:

• Mixed range types inside brackets, like s![[1..3, ..]] with a .. element: the array literal itself has a type mismatch (E0308).
• Unsupported index types, like 1u64..3: these fail on the From<_> for SliceInfoElem bound.

Tests

The new compile_fail,E0277 doc test checks that the bracketed form is rejected. Rustdoc only checks the error code on nightly. E0277 is what the compiler reports, as in the output above.

Checked locally on stable 1.99.1 (aarch64-apple-darwin): cargo clippy -F approx,serde,rayon, cargo test --doc, and cargo doc --no-deps --all-features. Not run locally: nightly cargo fmt (formatted by hand to match), MSRV 1.87, miri, nostd, cross, cargo-careful, BLAS.

🤖 Generated with Claude Code