cbcbox 2.910

Binary distribution of the CBC MILP solver (COIN-OR Branch and Cut)

cbcbox

cbcbox is a high-performance, self-contained Python distribution of the CBC MILP solver (COIN-OR Branch and Cut), built from the latest COIN-OR master branch.

On x86_64 (Linux, macOS, Windows) the wheel ships both a Haswell-optimised binary (AVX2/FMA, full -march=haswell ISA) for maximum speed and a generic build with runtime CPU dispatch for compatibility with any x86_64 machine — selected automatically. All dynamic dependencies (OpenBLAS, libgfortran, etc.) are bundled; no system libraries or separate installation steps are needed.

Highlights

• Haswell-optimised & generic builds — on x86_64 Linux, macOS, and Windows the wheel ships two complete solver stacks: a Haswell build (-O3 -march=haswell, OpenBLAS AVX2/FMA kernel) for maximum throughput, and a generic build (DYNAMIC_ARCH runtime dispatch) for compatibility with any x86_64 CPU. The best available variant is selected automatically at import time (see Build variants).

• Debug build — every wheel includes a debug variant compiled with -O1 -g (plus AddressSanitizer on macOS). Activate with CBCBOX_BUILD=debug to diagnose hard-to-find bugs with clean stack traces (see Debug build).

• Parallel branch-and-cut — built with --enable-cbc-parallel. Use -threads=N to distribute the search tree across N threads, giving significant speedups on multi-core machines for hard MIP instances.

• AMD fill-reducing ordering — SuiteSparse AMD is compiled in, enabling the high-quality UniversityOfFlorida Cholesky factorization for Clp's barrier (interior point) solver. AMD reordering produces much less fill-in on large sparse problems than the built-in native Cholesky, making barrier substantially faster. Activate with -barrier -cholesky UniversityOfFlorida (see barrier usage).

Performance (x86_64)

Auto-updated by CI after each successful workflow run. Single-threaded solve time — lower is better.

Single-threaded solve time across benchmark instances. Speedup factor shown above each pair. Lower is better.

Build variants

On x86_64 Linux, macOS, and Windows, the wheel ships two complete sets of binaries:

Variant	OpenBLAS kernel	Clp SIMD	Minimum CPU
generic	DYNAMIC_ARCH (runtime dispatch)	standard	any x86_64
avx2	HASWELL (256-bit AVX2/FMA)	-march=haswell -DCOIN_AVX2=4 (all Haswell ISA extensions + 4-double AVX2 layout)	Haswell (2013+)

At import time cbcbox automatically selects avx2 when it is available and the running CPU supports AVX2; otherwise it falls back to generic.

You can override this selection with the CBCBOX_BUILD environment variable:

# Force generic (portable) build
CBCBOX_BUILD=generic cbc mymodel.mps -solve -quit

# Force AVX2-optimised build (raises an error if not available)
CBCBOX_BUILD=avx2 cbc mymodel.mps -solve -quit

# Use debug build with AddressSanitizer (Linux/macOS) or -O1 -g (Windows)
CBCBOX_BUILD=debug cbc mymodel.mps -solve -quit

When CBCBOX_BUILD is set, a short summary of the selected build is printed to stdout on every call — useful for tagging experiment results:

[cbcbox] CBCBOX_BUILD=avx2
[cbcbox]   binary  : .../cbcbox/cbc_dist_avx2/bin/cbc
[cbcbox]   lib dir : .../cbcbox/cbc_dist_avx2/lib
[cbcbox]   libs    : libCbc.so.3, libClp.so.3, libopenblas.so.0

Non-x86_64 platforms (Linux aarch64, macOS arm64) ship generic and debug builds. CBCBOX_BUILD=avx2 will raise a RuntimeError on those platforms.

Debug build

Every wheel includes a debug build compiled with -O1 -g -fno-omit-frame-pointer. On macOS, AddressSanitizer (-fsanitize=address) is also enabled. Linux manylinux containers do not reliably provide libasan, so ASan is omitted there; Windows/MinGW does not support ASan either.

Use the debug build to reproduce and diagnose bugs: on macOS CBC will abort with a clear stack trace on memory errors; on all platforms reduced optimisation and full debug symbols make stack traces from crashes much more readable.

# Run with debug symbols (-O1 -g) — ASan also active on macOS
CBCBOX_BUILD=debug cbc problem.mps -solve -quit

import cbcbox, os
os.environ["CBCBOX_BUILD"] = "debug"
bin_path = cbcbox.cbc_bin_path()   # → .../cbc_dist_debug/bin/cbc

Supported platforms

Platform	Wheel tag
Linux x86_64	manylinux2014_x86_64
Linux aarch64	manylinux2014_aarch64
macOS arm64 (Apple Silicon)	macosx_11_0_arm64
macOS x86_64	macosx_10_9_x86_64
Windows AMD64	win_amd64

Installation

Note: cbcbox is now available on PyPI — pip install cbcbox. Pre-built wheel artifacts are also available from the CI runs (see below).

Installing from a pre-built wheel (recommended)

1. Go to the Actions tab of this repository.

2. Open the latest successful workflow run.

3. Download the artifact matching your platform:

   Artifact name	Platform
   cibw-wheels-Linux-X64	Linux x86_64
   cibw-wheels-Linux-ARM64	Linux aarch64
   cibw-wheels-macOS-ARM64	macOS Apple Silicon
   cibw-wheels-macOS-X64	macOS x86_64
   cibw-wheels-Windows-X64	Windows AMD64

4. Unzip the artifact and install the .whl file:

   pip install cbcbox-*.whl
   

Installing from PyPI

pip install cbcbox

Usage

Command line

After installation, CBC is available directly as the cbc command (pip installs the entry point into the environment's bin/ on Linux/macOS or Scripts/ on Windows, which is already on PATH):

cbc mymodel.lp -solve -quit
cbc mymodel.mps.gz -solve -quit
cbc mymodel.mps -seconds 60 -timem elapsed -solve -quit
cbc mymodel.mps -dualp pesteep -solve -quit

Alternatively, invoke via the Python module entry point:

python -m cbcbox mymodel.lp -solve -quit

CBC accepts LP, MPS and compressed MPS (.mps.gz) files. Pass -help for the full list of options, or -quit to exit after solving.

Parallel branch-and-cut

This build includes parallel branch-and-cut (--enable-cbc-parallel). Use -threads=N to distribute the search tree across N threads:

cbc mymodel.mps -threads=4 -solve -quit

Barrier (interior-point) solver

Clp's barrier solver can be faster than simplex for large LP relaxations. This build includes SuiteSparse AMD, which enables the high-quality UniversityOfFlorida Cholesky factorization — significantly reducing fill-in compared to the built-in native Cholesky:

# Solve LP relaxation with barrier + AMD Cholesky, then crossover to simplex basis
cbc mymodel.mps -barrier -cholesky UniversityOfFlorida -solve -quit

# Useful as a root-node strategy inside MIP (let CBC use simplex for B&B):
cbc mymodel.mps -barrier -cholesky UniversityOfFlorida -solve -quit

Without AMD, only -cholesky native (less efficient) is available.

Python API

The package exposes helpers to locate the installed files:

import cbcbox
import subprocess

# Path to the cbc binary (cbc.exe on Windows).
cbcbox.cbc_bin_path()
# e.g. '/home/user/.venv/lib/python3.13/site-packages/cbcbox/cbc_dist/bin/cbc'

# Directory containing the static and dynamic libraries.
cbcbox.cbc_lib_dir()
# e.g. '.../cbcbox/cbc_dist/lib'

# Directory containing the COIN-OR C/C++ headers.
cbcbox.cbc_include_dir()
# e.g. '.../cbcbox/cbc_dist/include/coin'

# Run CBC programmatically.
result = subprocess.run(
    [cbcbox.cbc_bin_path(), "mymodel.mps", "-solve", "-quit"],
    capture_output=True, text=True,
)
print(result.stdout)

What is built

The build pipeline compiles all components from source inside the CI runner, in the following order:

Component	Version / branch	Purpose
Cbc	master	Branch-and-cut MIP solver
Cgl	master	Cut generation library
Clp	master	Simplex LP solver (used as the MIP node relaxation)
Osi	master	Open Solver Interface
CoinUtils	master	Utility library (shared by all COIN-OR packages)
Nauty	2.8.9	Symmetry detection for MIP presolve
AMD (SuiteSparse v7.12.2)	v7.12.2	Sparse matrix fill-reducing ordering
OpenBLAS	v0.3.31	Optimised BLAS/LAPACK for LP basis factorisation

On x86_64 Linux, macOS, and Windows the entire stack is compiled twice: once for the generic variant (OpenBLAS DYNAMIC_ARCH=1) and once for the avx2 variant (TARGET=HASWELL, CXXFLAGS=-O3 -march=haswell -DCOIN_AVX2=4). AMD and Nauty are built only once (they are pure combinatorial code with no BLAS dependency) and reused by both COIN-OR variants.

All COIN-OR components are linked into both static (.a) and shared (.so / .dylib) libraries on Linux and macOS. On Windows only shared libraries (.dll) are produced — MinGW's autotools does not support building static and DLL simultaneously. The shared libraries are patched with self-relative RPATHs and bundled inside the wheel, making them directly usable via cffi or ctypes without any system installation.

Wheel contents

The wheel installs under cbcbox/ inside the site-packages directory. On x86_64 Linux, macOS, and Windows it contains two dist trees; other platforms contain only cbc_dist/:

cbc_dist/           ← generic build (all platforms)
cbc_dist_avx2/      ← AVX2-optimised build (x86_64 Linux/macOS/Windows)
├── bin/
│   ├── cbc           # CBC MIP solver binary  (cbc.exe on Windows)
│   └── clp           # Clp LP solver binary   (clp.exe on Windows)
├── lib/
│   ├── libCbc.so / libCbc.dylib / libCbc.dll  # CBC solver
│   ├── libCbcSolver.so ...
│   ├── libClp.so ...                          # Clp LP solver
│   ├── libCgl.so ...                          # Cut generation
│   ├── libOsi.so ...                          # Solver interface
│   ├── libOsiClp.so ...                       # Clp OSI binding
│   ├── libOsiCbc.so ...                       # CBC OSI binding (where available)
│   ├── libCoinUtils.so ...
│   ├── libopenblas.so / .dylib / .dll         # OpenBLAS BLAS/LAPACK
│   ├── pkgconfig/                             # .pc files for all libraries
│   └── <bundled runtime shared libs>          # Platform-specific — see below
└── include/
    ├── coin/      # COIN-OR headers (CoinUtils, Osi, Clp, Cgl, Cbc)
    ├── nauty/     # Nauty headers
    └── *.h        # SuiteSparse / AMD headers

Bundled dynamic libraries

Because the static COIN-OR libraries link to OpenBLAS, which in turn links to the Fortran runtime, the following shared libraries are bundled inside the wheel and their paths are rewritten so no system installation is required.

Linux (lib/ directory, RPATH set to $ORIGIN)

Library	Description
libopenblas.so.0	OpenBLAS BLAS/LAPACK
libgfortran.so.5	GNU Fortran runtime
libquadmath.so.0	Quad-precision math (dependency of libgfortran)

macOS (lib/ directory, install names rewritten to @rpath/)

Library	Description
libopenblas.dylib	OpenBLAS BLAS/LAPACK
libgfortran.5.dylib	GNU Fortran runtime
libgcc_s.1.1.dylib	GCC runtime
libquadmath.0.dylib	Quad-precision math

Windows (bin/ directory, DLLs placed next to the executable)

Library	Description
libopenblas.dll	OpenBLAS BLAS/LAPACK
libgfortran-5.dll	GNU Fortran runtime
libgcc_s_seh-1.dll	GCC SEH runtime
libquadmath-0.dll	Quad-precision math
libstdc++-6.dll	C++ standard library (MinGW64)
libwinpthread-1.dll	POSIX thread emulation

CI / build pipeline

Wheels are built and tested automatically via GitHub Actions using cibuildwheel. The workflow (.github/workflows/wheel.yml) runs on five separate runners:

Runner	Produces
ubuntu-latest	manylinux2014_x86_64 wheel
ubuntu-24.04-arm	manylinux2014_aarch64 wheel
macos-15	macosx_11_0_arm64 wheel
macos-15-intel	macosx_10_9_x86_64 wheel
windows-latest	win_amd64 wheel

After each wheel is built, the test suite in tests/ is run against the installed wheel to verify correctness.

Integration tests

The test suite (pytest) solves fifteen MIP instances and checks the optimal objective values, in both single-threaded and parallel (3-thread) modes. On x86_64 Linux, macOS, and Windows each test is run twice — once against the generic binary and once against the avx2 binary — and a side-by-side performance comparison is recorded:

Instance	Expected optimal	Time limit
pp08a.mps.gz	7 350	2000 s
sprint_hidden06_j.mps.gz	130	2000 s
air03.mps.gz	340 160	2000 s
air04.mps.gz	56 137	2000 s
air05.mps.gz	26 374	2000 s
nw04.mps.gz	16 862	2000 s
mzzv11.mps.gz	−21 718	2000 s
trd445c.mps.gz	−153 419.078836	2000 s
nursesched-sprint02.mps.gz	58	2000 s
stein45.mps.gz	30	2000 s
neos-810286.mps.gz	2 877	2000 s
neos-1281048.mps.gz	601	2000 s
j3050_8.mps.gz	1	2000 s
qiu.mps.gz	−132.873136947	2000 s
gesa2-o.mps.gz	25 779 856.3717	2000 s

Time limits are generous to avoid false failures on slow CI runners.

Publishing to PyPI

Note: cbcbox is not yet registered on PyPI. When ready, trigger the workflow manually and select pypi (or testpypi) in the Publish input. Trusted Publisher (OIDC) authentication is used — no API tokens are stored as secrets.

Performance results

Auto-updated by CI after each successful workflow run.

Summary

Geometric mean solve time (seconds) across all test instances.

1 thread

Platform	generic (s)	avx2 (s)	avx2 speedup
Darwin x86_64	67.84	23.09	2.94×
Darwin arm64	56.25	—	—
Windows AMD64	61.10	20.01	3.05×

3 threads

Platform	generic (s)	avx2 (s)	avx2 speedup
Darwin x86_64	50.68	21.99	2.30×
Darwin arm64	38.06	—	—
Windows AMD64	56.17	21.21	2.65×

Per-instance results

pp08a.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	5.16	13.00	0.40×
Darwin x86_64	generic	10.72	7.76	1.38×
Darwin arm64	generic	15.16	15.58	0.97×
Windows AMD64	avx2	5.01	9.58	0.52×
Windows AMD64	generic	12.66	19.82	0.64×

sprint_hidden06_j.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	55.51	57.49	0.97×
Darwin x86_64	generic	188.25	201.09	0.94×
Darwin arm64	generic	182.51	123.93	1.47×
Windows AMD64	avx2	59.08	62.13	0.95×
Windows AMD64	generic	252.61	235.90	1.07×

air03.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	1.87	2.08	0.90×
Darwin x86_64	generic	6.44	7.49	0.86×
Darwin arm64	generic	5.79	4.01	1.45×
Windows AMD64	avx2	2.48	2.59	0.96×
Windows AMD64	generic	5.88	6.05	0.97×

air04.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	58.54	42.58	1.37×
Darwin x86_64	generic	118.34	76.36	1.55×
Darwin arm64	generic	136.97	73.22	1.87×
Windows AMD64	avx2	33.37	28.11	1.19×
Windows AMD64	generic	156.20	88.02	1.77×

air05.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	26.66	20.40	1.31×
Darwin x86_64	generic	59.29	51.91	1.14×
Darwin arm64	generic	64.25	36.62	1.75×
Windows AMD64	avx2	18.25	14.16	1.29×
Windows AMD64	generic	58.30	51.63	1.13×

nw04.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	13.79	15.72	0.88×
Darwin x86_64	generic	41.79	43.26	0.97×
Darwin arm64	generic	40.06	35.31	1.13×
Windows AMD64	avx2	16.05	16.72	0.96×
Windows AMD64	generic	57.63	54.43	1.06×

mzzv11.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	139.30	106.96	1.30×
Darwin x86_64	generic	567.19	402.04	1.41×
Darwin arm64	generic	294.85	159.32	1.85×
Windows AMD64	avx2	120.97	176.20	0.69×
Windows AMD64	generic	260.16	310.18	0.84×

trd445c.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	119.23	116.71	1.02×
Darwin x86_64	generic	251.27	239.97	1.05×
Darwin arm64	generic	221.60	171.13	1.29×
Windows AMD64	avx2	106.86	106.49	1.00×
Windows AMD64	generic	244.50	241.21	1.01×

nursesched-sprint02.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	36.95	37.11	1.00×
Darwin x86_64	generic	102.12	107.08	0.95×
Darwin arm64	generic	97.01	92.82	1.05×
Windows AMD64	avx2	27.40	27.26	1.01×
Windows AMD64	generic	118.73	90.77	1.31×

stein45.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	10.18	9.23	1.10×
Darwin x86_64	generic	28.15	18.84	1.49×
Darwin arm64	generic	22.16	10.52	2.11×
Windows AMD64	avx2	8.66	6.27	1.38×
Windows AMD64	generic	27.00	18.95	1.42×

neos-810286.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	14.98	14.93	1.00×
Darwin x86_64	generic	46.87	46.77	1.00×
Darwin arm64	generic	29.20	27.42	1.07×
Windows AMD64	avx2	13.39	13.58	0.99×
Windows AMD64	generic	36.56	47.59	0.77×

neos-1281048.mps.gz

Platform	Build	1 thread (s)	3 threads (s)	parallel speedup
Darwin x86_64	avx2	21.28	9.47	2.25×
Darwin x86_64	generic	130.03	15.72	8.27×
Darwin arm64	generic	43.34	17.28	2.51×
Windows AMD64	avx2	13.98	18.56	0.75×
Windows AMD64	generic	36.80	23.01	1.60×

NAQ — Never Asked Questions

Why not benchmark on the full MIPLIB 2017 library?

Several practical constraints shape the benchmark set:

1. CI time limits. GitHub Actions enforces a 6-hour wall-clock limit per job. The full MIPLIB 2017 collection contains ~1 065 instances, many of which take hours even on fast hardware. Including all of them would make every CI run time out before producing any useful measurements.

2. Comparing apples to apples requires instances solved to optimality. If some instances are only solved within a time limit (i.e., a gap > 0 %), a meaningful performance comparison must account for both solve time and solution quality simultaneously. This greatly complicates analysis and makes plots harder to interpret. Restricting to instances that CBC reliably solves to proven optimality keeps the comparison clean: a single elapsed-time number per instance is all that is needed.

3. The instance set is intentionally biased toward set packing / covering / partitioning structure. Most instances in the benchmark (pp08a, sprint_hidden06_j, nw04, mzzv11, nursesched-sprint02, air0x, trd445c) contain large blocks of set packing, covering, or partitioning constraints. This structure arises naturally in applications such as crew scheduling, nurse scheduling, vehicle routing, and cutting stock — exactly the domain where column generation is most valuable. Since the benchmark focuses on this problem class rather than providing a general-purpose solver survey, it is a specially interesting use case.

License

CBC and all COIN-OR components are distributed under the Eclipse Public License 2.0. OpenBLAS is distributed under the BSD 3-Clause licence. SuiteSparse AMD is distributed under the BSD 3-Clause licence. Nauty is distributed under the Apache 2.0 licence.