Expand description
§Chapter 12.0.2 — Dense views
§Purpose
The caller’s data — an R matrix, a NumPy array in Fortran order, a
block inside a larger array — is borrowed, not copied: MatRef and
MatMut are column-major views with a leading dimension (decision S3
of D-0127). The owned type of the crate’s outputs is
DenseBlock of xingo-sparse (ld = rows).
§Notation
A view is rows × cols with leading dimension ld ≥ rows: a_ij = data[i + j·ld]. Column j is the contiguous slice `data[j·ld .. j·ld
- rows]
. A C-order NumPy array is the view of its *transpose* (a rule forxingo-ffi`).
§Algorithm V (checked construction)
V1. [Leading dimension.] If ld < rows, InvalidArgument.
V2. [Extent.] If cols = 0 the extent is 0; otherwise need = ld·(cols − 1) + rows, computed with checked arithmetic (InvalidArgument on
usize overflow).
V3. [Length.] If data.len() < need, Dimension { what: "data", expected: need, found: data.len() }. Longer slices are accepted
(the tail is not part of the view).
Finiteness is not checked here, but by the consumers (each reports
the column-major position i + j·rows).
§Invariant
(W1) ld ≥ rows and data.len() ≥ need for every live view, so every
accessor below indexes in bounds; the accessors that take indices
return None (or an error) out of range instead of panicking.
Structs§
- MatMut
- The mutable counterpart of
MatRef. - MatRef
- A borrowed column-major
rows × colsmatrix with leading dimensionld ≥ rows:a_ij = data[i + j·ld](the LAPACK/R layout; a C-order NumPy array is the view of its transpose, a rule forxingo-ffi).data.len() ≥ ld·(cols − 1) + rowswhencols > 0.
Functions§
- extent 🔒
- Algorithm V: the extent
needof a view, or the error.