nalgebra

Release nalgebra-glm, -lapack, and -sparse

Added

• Add the Bunch–Kaufman LDL decomposition (LBLT) via the new Matrix::lblt method #1591.
• Add OneNorm and the Matrix::one_norm method computing the induced matrix 1-norm (maximum absolute column sum) #1591.
• Add the convert-glam033 feature to enable conversion from/to types from glam v0.33.

Changed

• Bumped MSRV to 1.89.0.
• Updated simba dependency to 0.10.
• Updated rand dependency to 0.10 and rand_distr to 0.6.
• Removed the convert-glamXXX features for all glam versions prior to 0.30 #1598.
• Clarified norm documentation: EuclideanNorm is the Frobenius norm for matrices, LpNorm/UniformNorm are entrywise, not induced matrix norms #1591.
• nalgebra-lapack: updated thiserror to 2.0.

Fixed

• Fix the T: Sync bound on impl Send for ViewStorage so view storages can be sent across threads when T: Send #1581.

Added

• Add convert-glam031 and convert-glam032 features for conversion from/to glam v0.31 and v0.32 #1597.

Fixed

• Fix SymmetricEigen routine #1210.
• Fix matrix parsing grammar to accept numbers without leading zeros, e.g. ".45" #1578.
• Fix build with glam in no_std environments #1555.
• Fix rustdoc warnings #1511.
• Implement absolute threshold for early deflation in Schur algorithm #1565.
• Assert matrix shapes for Matrix::abs_diff_eq #1568.

Release v0.34.1 (#1546)

Added

• Added encase feature, providing encase trait implementations for nalgebra types.

Added

• Add the convert-glam030 feature to enable conversion from/to types from glam v0.30.
• Add the defmt cargo feature that enables derives of defmt::Format for all no-std types.

Changed

• Bumped MSRV to 1.87.0.
• Updated rand dependency to 0.9.0.
• Renamed associated const DimName::USIZE to DimName::DIM.
• Moved to Rust 2024 edition.
• Several methods are now const whenever possible. See details in #1522.
• Features for conversion from/to types from glam (such as convert-glam029) no longer enable default features for glam, allowing use in no_std environments.

Fixed

• Fix infinite loop when attempting to take the Schur decomposition of a 0 matrix.

Co-authored-by: Willow Black [email protected]

Release v0.33.2 (#1457)

Added

• Add the convert-glam029 feature to enable conversion from/to types from glam v0.29.

Added

• Add implementations of bytemuck traits for isometries and similarities.
• Implement AsRef<[T]> for matrices with contiguous storage.
• Enable the num-complex/bytemuck feature when the convert-bytemuck feature is enabled.

Fixed

• Fix a memory leak in Matrix::generic_resize.
• Fix glm::is_null to check the vector magnitude instead of individual components.
• Ensure that inverting a 4x4 matrix leaves it unchanged if the inversion fails.

Added

• Add the glam-0.28 feature to enable conversion from/to types from glam v0.28.
• Add a stack! macro for concatenating matrices. See #1375.

Modified

• The cuda feature has been removed, as the toolchain it depends on is long abandoned.
• Update to simba 0.9. See the changelog of simba for details.
• Update the nalgebra-macros crate to syn 2.0.
• Remove the scalar type T from the Allocator trait parameters. Instead of Allocator<T, R, C>, use the simpler Allocator<R, C>.

Added

• Add the glam-0.27 feature to enable conversion from/to types from glam v0.27.

Added

• Add the glam-0.25 feature to enable conversion from/to types from glam v0.25.

Modified

• Statically sized matrices are now serialized as tuples to match how serde serialized plain arrays.
• Don’t require Scalar for matrix PartialEq and Eq.

Added

• Allow trailing punctuation in macros vector!, matrix!, point!, etc.
• Add the methods Matrix1::as_scalar, ::as_scalar_mut, ::to_scalar, ::into_scalar.
• Add Rotation3::euler_angles_ordered, a generalized euler angles calculation.
• Add the glam-0.24 feature to enable conversion from/to types from glam v0.24.
• Add the glam-0.25 feature to enable conversion from/to types from glam v0.25.
• Add the lerp method to points.
• Implement Clone for MatrixIter.

Fixed

• Fixed severe catastrophic cancellation issue in variance calculation.

Added

• Add the glam-0.23 to enable conversion from/to type from glam v0.23.

Modified

• Updated nalgebra-macros to use the new Dyn, avoiding macro-generated deprecation warnings.

Modified

• Renamed all MatrixSlice types to MatrixView. In general all occurrences of the world Slice or slice have been replaced by View or view.
• Deprecated all the types involving Slice in its name, in favor of the word View.
• Make it so that most nalgebra objects archive as themselves (when using rkyv for serialization).
• Renamed Dynamic to Dyn and make Dyn a tuple struct.

Added

• Add Cholesky::ln_determinant to compute the natural logarithm of the determinant of a matrix decomposed with Cholesky. This can be more numerically stable than computing the determinant itself when very small and/or large values are involved.
• Added new methods Matrix::as_view and Matrix::as_view_mut, which are very useful when working with view-based APIs.
• Added parallel column iterator using rayon: Matrix::par_column_iter and Matrix::par_column_iter_mut. The rayon feature must be enabled to access these methods.
• Implement ReshapableStorage for matrix slices (only for unit strides at the moment).
• Add U0, U1, … constants alongside the U0, U1, … types. This lets us write U4 instead of U4::name() or Const::<4> when we need const dimensions.

Fixed

• Fixed the implementation of Rotation3::euler_angles to return the angles in the documented order (roll, pitch, yaw).

Added

• Add a convert-glam022 feature to enable conversion between nalgebra and glam v0.22.

Added

• Add Matrix::try_cast to attempt casting the inner scalar types when that cast may fail.

Fixed

• Fixed the usage of CheckBytes with rkyv.

Modified

• Use #[inline] on the Dim implementation for Const to improve opt-level 1 performance.
• Make the Point::new constructions const-fn.

Added

• Add UnitVector::cast to change the underlying scalar type.

Modified

• Improve performances of multiplication of two sparse matrices.

Added

• Add Matrix::from_row_iterator to build a matrix from an iterator yielding components in row-major order.
• Add support for conversion from/to types of glam 0.21.
• nalgebra-sparse: add support for the matrix-market export of sparse matrices.
• nalgebra-lapack: add a GE for solving the generalized eigenvalues problem.

Fixed

• Fix Rotation3::from_matrix and UnitQuaternion::from_matrix when the input matrix is already a valid rotation matrix.

Breaking changes

• Switch to cust 0.3 (for CUDA support).
• Switch to rkyv 0.7
• Remove support for serialization based on abomonation.
• Remove support for conversions between nalgebra types and glam 0.13.

Modified

• The aliases for Const types have been simplified to help rust-analyzer.

Added

• Add TryFrom conversion between UnitVector2/3/4 and glam’s Vec2/3/4.
• nalgebra-sparse: added support for serialization of sparse matrices with serde.
• nalgebra-sparse: add a CSC matrix constructor from unsorted (but valid) data.
• nalgebra-lapack: add generalized eigenvalues/eigenvectors calculation + QZ decomposition.

Fixed

• Improve stability of SVD.
• Fix slerp for UnitComplex.

Added

• Add conversion from/to types of glam 0.19 and 0.20.

Breaking changes

• The Dim trait is now marked as unsafe.
• The Matrix::pow and Matrix::pow_mut methods only allow positive integer exponents now. To compute negative exponents, the user is free to invert the matrix before calling pow with the exponent’s absolute value.
• Remove the Bounded requirement from RealField. Replace it by methods returning Option<Self> so that they can still be implemented by unbounded types (by returning None).
• The ComplexField trait derives from FromPrimitive again. We can actually keep this because all its methods return Option<Self>, meaning that it could be implemented by any type.

Modified

• Use more concise debug impls for matrices and geometric transformation types.
• The singular values computed by the SVD are now sorted in increasing order by default. Use SVD::new_unordered instead to reproduce the older behavior without the sorting overhead.
• The UnitDualQuaternion::sclerp method will no longer panic when given two equal rotations.
• The Matrix::select_rows and Matrix::select_columns methods no longer require the matrix components to implement the trait Zero.
• The Matrix::pow and Matrix::pow_mut methods will now also work with integer matrices.

Added

• Added the conversion trait From<Vec<T>> and method from_vec_storage for RowDVector.
• Added implementation of From and Into for converting between nalgebra types and types from glam 0.18. These can be enabled by enabling the convert-glam018 cargo features.
• Added the methods Matrix::product, ::row_product, ::row_product_tr, and ::column_product to compute the product of the components, rows, or columns, of a single matrix or vector.
• The Default trait is now implemented for most geometric types: Point, Isometry, Rotation, Similarity, Transform, UnitComplex, and UnitQuaternion.
• Added the Scale geometric type for representing non-uniform scaling.
• Added Cholesky::new_with_substitute that will replace diagonal elements by a given constant whenever Cholesky meets a non-definite-positiveness.
• Re-added the conversion from a vector/matrix slice to a static array.
• Added the cuda feature that enables the support of rust-cuda for using nalgebra features with CUDA kernels written in Rust.
• Added special-cases implementations for the 2x2 and 3x3 SVDs for better accuracy and performances.
• Added the methods Matrix::polar, Matrix::try_polar, and SVD::to_polar to compute the polar decomposition of a matrix, based on its SVD.
• nalgebra-sparse: provide constructors for unsorted but otherwise valid data using the CSR format.
• nalgebra-sparse: added reading MatrixMarked data files to a sparse CooMatrix.

Fixed

• Fixed a potential unsoundness with matrix.get(i) and matrix.get_mut(i) where i is an usize, and matrix is a matrix slice with non-default strides.
• Fixed potential unsoundness with vector.perp where vector isn’t actually a 2D vector as expected.
• Fixed linkage issue with nalgebra-lapack: the user of nalgebra-lapack no longer have to add extern crate lapack-src to their main.rs.
• Fixed the no-std build of nalgebra-glm.
• Fix the pow and pow_mut functions (the result was incorrect for some exponent values).

This updates all the dependencies of nalgebra to their latest version, including:

• rand 0.8
• proptest 1.0
• simba 0.4

New crate: nalgebra-sparse

Alongside this release of nalgebra, we are releasing nalgebra-sparse: a crate dedicated to sparse matrix computation with nalgebra. The sparse module of nalgebraitself still exists for backward compatibility, but it will be deprecated soon in favor of the nalgebra-sparse crate.

Added

• Add UnitDualQuaternion, a dual-quaternion with unit magnitude which can be used as an isometry transformation.
• Add UDU::new() and matrix.udu() to compute the UDU factorization of a matrix.
• Add ColPivQR::new() and matrix.col_piv_qr() to compute the QR decomposition with column pivoting of a matrix.
• Add from_basis_unchecked to all the rotation types. This builds a rotation from a set of basis vectors (representing the columns of the corresponding rotation matrix).
• Add Matrix::cap_magnitude to cap the magnitude of a vector.
• Add UnitQuaternion::append_axisangle_linearized to approximately append a rotation represented as an axis-angle to a rotation represented as an unit quaternion.
• Mark the iterators on matrix components as DoubleEndedIter.
• Re-export simba::simd::SimdValue at the root of the nalgebra crate.