Osqp Versions Save

The v1.0.0.beta1 pre-release of osqp adds CI-generated build artifacts for several platforms.

First beta release of OSQP v1.0

New features:

• Introduced new linear algebra backend system allowing compute framework to be changed at compile time.
• Merged cuOSQP project into main OSQP project (inside algebra/cuda directory).
• Introduced an Intel MKL-based algebra backend using the MKL sparse BLAS API, Vector Math Library. This backend contains both the Pardiso solver and an RCI conjugate gradient implementation.
• Added code generation capabilities to the C-level API (note, only problem export is in the C API, no file copying is done).
• Added initial adjoint derivative computation to the C-level API.

Main changes:

• Updated QDLDL to 0.1.7.
• Changed QDLDL to be included through CMake FetchContent instead of a git submodule.
• The MKL Pardiso solver is only available with the MKL backend.
• CMake installs CMake config files for consuming applications to use.
• All of OSQP's API is contained inside the public include files (there should be no need for users to include anything in private).
• All OSQP functions, defines and types are prefixed with OSQP (in some capitalization) to namespace the API.

Developer-centric changes:

• Test suite switched to Catch2 and now incorporates modern C++ for memory management and organization.
• ASAN flags integrated into main OSQP CMake build system.

• Fix MKL function prototypes (required for CRAN compilation) (PR https://github.com/osqp/osqp/pull/487)
• Use a constant interval for adaptive rho when in embedded=2 mode (PR https://github.com/osqp/osqp/pull/347)
• Include version.h in the OSQP installed headers (Fixes https://github.com/osqp/osqp/issues/323)
• Switch unit testing to use Catch2
• Switch to GitHub actions CI system
• Switched binary distribution from bintray to GitHub releases
• Various documentation fixes and improvements

• Fix segfault python multithreading
• Compatibility python 3.9
• Updated QDLDL to version 0.1.5
• Drop Python 2.7 support

• Added meaningful return values to internal functions. Changed syntax of osqp_setup function. It now returns an exitflag.
• osqp_setup function requires P to be upper triangular. It returns a nonzero exitflag otherwise.
• Custom memory allocators via cmake and the configure file.
• Changed interfaces to linsys solver functions. The solve function now computes (x_tilde,z_tilde) instead of (x_tilde,nu). This allows to implement custom linear system solvers (also indirect).
• Added solve function in Python interface that performs setup solve and cleanup for you directly and disables GIL.
• Improved code generation folder structure.

• Added update_time to the info structure.
• Fixed #101.

• Updated QDLDL to version 0.1.3.
• Added check for nonconvex cost function (non-positive semidefinite P) after factorization is performed.
• Added complete sources distribution on bintray.com (including QDLDL).

• Removed SuiteSparse LDL in favor of QDLDL
• Added check for nonconvex cost function (P not positive semidefinite).
• Static library libosqpstatic now renamed as libosqp.

• Fixed #62.
• Moved interfaces to separate repositories
• Fixed #54.
• Changes to support Matlab 2018a
• Added support for new interface in R

Features

• Added time_limit option
• Added CUTEst interface
• Add "install" CMake target and installation of CMake configuration files

Interface changes

• Changed set_default_settings to osqp_set_default_settings and brought function to main API header osqp.h
• Allow eps_rel=0 #40

Bugs

• Fixed bug in upper triangular P extraction. Now the solver can accept both complete P matrix or just the upper triangular part.
• Fixed #33
• Fixed #34
• Fixed bug when calling osqp_solve or osqp_cleanup after failed linear system initialization
• Fixed potential name conflict with SCS 47
• Fixed #49