v9.12

[Mizux]

Platforms

• This will be the last release to support Python3.8.
• Add Python 3.13 support.
• Add muslinux wheel package.
• This will be the last release to support Java 1.8

Builds

• dependencies (protobuf, abseil, re2, highs) are now seperated shared libraries
• Windows libraries are now dll

Dependencies

C++

• Protobuf 29.3
• CoinUtils: 2.11.12
• Osi: 0.108.11
• Clp: 1.17.10
• Cgl: 0.60.9
• Cbc: 2.10.12
• HiGHS: 1.9.0
• Scip: v920
• benchmark: 1.9.1

Graph

• Old graph classes (EbertGraph, StarGraph, ForwardEbertGraph, ForwardStarGraph) have been removed.
  Use ReverseArcListGraph<> and ListGraph<> instead. Those are conceptually similar, but typically perform better.
  APIs are mostly compatible.

CP-SAT

• Improve the no_overlap_2d layer (propagation, presolve, cuts)
• Add RandomHalfBranching search heuristics
• Improve Cumulative propagation and cuts
• Rewrote the hint transformation during presolve. Feasible and complete hints should now be preserved after presolve
• Rewrite and optimize the linear expression part of the python layer
• Improve shared tree workers, implement job affinity with workers
• Add scheduling samples
• Support 1-var affine expressions in the Automaton constraint
• Support 1-var affine expressions in the Element constraint
• Speed up LNS
• Tens of fixes for bugs found by fuzzer

Linear Solver

• Add WriteModelToMpsFile to MPSolver in Java, Python and .NET

Math Opt

• Add Xpress support.

Model Builder

• Speedup of the python expression part.
• Add HiGHS support.

SCIP

• This is likely the last release of SCIP build with glop
• We will transition to SCIP with SOPLEX builds
• This is already the case for the Bazel build

Known Issues

• MathOpt: python_math_opt_solver_test failed on macOS MathOpt: python_math_opt_solver_test failed on macOS #4460
• cmake: MacOS Intel based python wheel do not contains mypy files cmake:python: stubgen timout on macos intel #4540
• cmake: cxx_graph_shortest_paths_test (Timeout) cmake: cxx_graph_shortest_paths_test (Timeout) #4539
• CP-SAT: python add_abs_equality broken OPTIMAL status in v9.11.4210 but INFEASIBLE status in v9.12.4544 #4568

Full Changelog: v9.11...v9.12

---

This discussion was created from the release v9.12.

[jmarca]

Thanks for this! I'm seeing a significant speedup for some of my CP-SAT problem sets.

[jmarca]

The specific application is a routing problem that uses the circuit constraints in CP-SAT. The problem is more like resource allocation, matching pickup and delivery of items with truck capacity, crew availability, driver availability, venue and warehouse operating hours, and so on. Too complicated for the routing solver. A known largish problem I use for testing takes about 85 seconds with 9.11, and about 56 seconds 9.12 to converge to optimal, on my laptop with 8 workers (8 threads cpu).  James