islpy-barvinok 2025.2.5rc1

islpy built with Barvinok and vendored native libs (Linux/macOS). Falls back to source build if wheel unavailable.

islpy-barvinok

Prebuilt wheels and from-source builds of islpy with Barvinok enabled, with native libraries vendored on Linux and macOS.

This project is a wrapper/builder: it fetches the upstream islpy sdist at a pinned version, builds Barvinok/NTL/ISL via a reproducible script, builds islpy with USE_BARVINOK=ON, and publishes wheels. It keeps the import name as islpy for compatibility, while the distribution name on PyPI is islpy-barvinok.

Supported platforms: Linux x86_64 (manylinux) and macOS 11+ (x86_64 and arm64).

Upstream project: inducer/islpy. This repo tracks upstream islpy releases on PyPI and rebuilds with Barvinok enabled.

Usage

• Install: pip install islpy-barvinok
• Import:

import islpy
islpy.Set("[N] -> {[x, y] : x >= 0 and y >= 0 and x <= N and y <= N}").card()

How the wheels are built (overview)

• Build dependencies into a local prefix (default build/prefix):
  • GMP, NTL, ISL, Barvinok (from upstream sources)
• Build islpy against the prefix with:
  • -D USE_SHIPPED_ISL=OFF -D USE_SHIPPED_IMATH=OFF -D USE_BARVINOK=ON
  • -D ISL_INC_DIRS:LIST=<prefix>/include
  • -D ISL_LIB_DIRS:LIST=<prefix>/lib
• Vendor native libs into wheels:
  • Linux: auditwheel repair
  • macOS: delocate-wheel

Building from source (PEP 517)

If a prebuilt wheel is unavailable for your platform, pip will build from source using a custom backend that runs scripts/build_all.sh. You will need common C/C++ build tools and autotools installed system-wide:

• macOS: brew install autoconf automake libtool pkg-config cmake ninja
• Debian/Ubuntu: sudo apt-get install -y build-essential autoconf automake libtool pkg-config cmake ninja-build

Then:

pip install --no-binary=:all: islpy-barvinok

The build downloads sources for GMP, NTL, ISL, and Barvinok, builds them into a local prefix under build/, builds islpy against that prefix with Barvinok enabled, and produces a wheel that pip installs. On macOS, native libraries may be vendored using delocate if available.

Releases and automation

• Upstream tracking: a scheduled workflow checks PyPI for new islpy releases and opens a PR bumping the pinned version.
• Tagging: when such a PR is merged into main, another workflow tags the repo with v<version>.
• Publishing: tags trigger CI to build wheels on Linux and macOS and publish to PyPI via Trusted Publishing (OIDC).

Development

• Single-script build:
  • uv run bash scripts/build_all.sh
  • Wheels in build/wheelhouse and build/wheelhouse-repaired.