Do not panic on empty ranges. (#257)

candle-core/src/indexer.rs

@@ -42,7 +42,7 @@ impl Tensor {
                         Bound::Excluded(n) => *n,
                         Bound::Unbounded => dims[i],
                     };
-                    let out = x.narrow(current_dim, start, stop - start)?;
+                    let out = x.narrow(current_dim, start, stop.saturating_sub(start))?;
                     current_dim += 1;
                     out
                 }

candle-core/tests/indexing_tests.rs

@@ -58,6 +58,19 @@ fn range_index() -> Result<()> {
     let result = tensor.i(..=1)?;
     assert_eq!(result.dims(), &[2, 3]);
     assert_eq!(result.to_vec2::<u32>()?, &[[0, 1, 2], [3, 4, 5]]);
+
+    // Empty range
+    let result = tensor.i(1..1)?;
+    assert_eq!(result.dims(), &[0, 3]);
+    let empty: [[u32; 3]; 0] = [];
+    assert_eq!(result.to_vec2::<u32>()?, &empty);
+
+    // Similar to PyTorch, allow empty ranges when the computed length is negative.
+    #[allow(clippy::reversed_empty_ranges)]
+    let result = tensor.i(1..0)?;
+    assert_eq!(result.dims(), &[0, 3]);
+    let empty: [[u32; 3]; 0] = [];
+    assert_eq!(result.to_vec2::<u32>()?, &empty);
     Ok(())
 }