Setup C++ Infra in GiGL

[mkolodner-sc]

Scope of work done

C++ Infrastructure for GiGL

This PR adds the foundational C++ infrastructure to GiGL — everything needed to write, build, lint, and test C++ extensions alongside the existing Python codebase, and the release pipeline to ship them.

Two-wheel model

GiGL is now distributed as two wheels installed together:

• gigl — pure Python (same wheel for CPU and CUDA users)
• gigl-core — compiled C++/CUDA pybind11 extensions, ABI-bound to the torch variant. Currently contains no C++ sources; the package builds an empty wheel now and gets populated as extensions are added.

Both wheels are versioned and released together. gigl-core is a dependency of gigl and is resolved automatically from the same GCP registry — no separate install step for users.

Build system

gigl-core/ is a standalone scikit-build-core + CMake package. CMakeLists.txt auto-discovers all csrc/python_*.cpp (and python_*.cu) files as pybind11 extension modules — adding a new extension requires only dropping source files, no CMake edits. pybind11 is fetched via FetchContent so it does not need to be pre-installed. Torch is located by asking the target Python directly (torch.utils.cmake_prefix_path).

no-build-isolation-package = ["gigl-core"] in pyproject.toml disables build isolation for gigl-core so cmake can find torch (not on PyPI) from the ambient venv when real extensions are added. The build-backend dependency group provides scikit-build-core as the PEP 517 build backend in both dev and Docker environments.

Release pipeline

release.yml builds and publishes both wheels via a matrix build — one CPU runner, one GPU runner — to two GCP Artifact Registry repos (gigl for CPU, gigl-cu128 for CUDA 12.8). RELEASING.md documents the two-workflow release process: Create Release (version bump, branch, tag, PR) followed by Release GiGL (wheel build and publish).

C++ tooling

• Formatting — .clang-format establishes style conventions. make format_cpp formats in-place.
• Linting — .clang-tidy config. make check_lint_cpp runs clangd --check in parallel (faster than clang-tidy due to preamble caching). make fix_lint_cpp applies auto-fixable violations via clang-tidy --fix (intentionally separate from make format since it rewrites logic, not just style).
• Dependency installation — requirements/install_cpp_deps.sh installs clang-format-15, clang-tidy-15, clangd-15, clang++-15, and libstdc++-12-dev on Linux. (clang++-15 requires libstdc++-12-dev on Ubuntu 22.04 to find standard headers.)
• Style guide — docs/cpp_style_guide.md documents formatting rules, linting rationale, and disabled checks.

C++ unit tests

gigl-core/tests/ with GoogleTest (via FetchContent). Any *_test.cpp or *_test.cu file is automatically discovered and registered — no CMake edits needed to add a test. Run via make unit_test_cpp; included in make unit_test. A stamp file (.cache/cpp_tests/.configured) avoids re-running cmake configure when nothing changed; CONFIGURE_DEPENDS auto-invalidates when source files are added or removed.

Where is the documentation for this feature?: N/A

Did you add automated tests or write a test plan?

Updated Changelog.md? NO

Ready for code review?: NO

[github-actions]

GiGL Automation

@ 23:19:08UTC : Starting to build base images for CUDA and CPU.
This may take a while, please be patient.
The images will be pushed to the GCP Artifact Registry.
Once done, the workflow will update the gigl/dep_vars.env file with the new image names.

[github-actions]

GiGL Automation

@ 23:22:33UTC : Starting to build base images for CUDA and CPU.
This may take a while, please be patient.
The images will be pushed to the GCP Artifact Registry.
Once done, the workflow will update the gigl/dep_vars.env file with the new image names.

[github-actions]

GiGL Automation

@ 23:32:11UTC : Built and pushed new images:

• CUDA base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cuda-base:b598f3d72eee47f5513dcb39460944459a0a012f.108.1
• CPU base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cpu-base:b598f3d72eee47f5513dcb39460944459a0a012f.108.1
• Dataflow base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-dataflow-base:b598f3d72eee47f5513dcb39460944459a0a012f.108.1
• Builder image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/gigl-base-images/gigl-builder:b598f3d72eee47f5513dcb39460944459a0a012f.108.1

Updated gigl/dep_vars.env with new image names.
Updated .github/cloud_builder/run_command_on_pr_cloud_build.yaml with new builder image.

[svij-sc]

Thanks!
Once the subsequent PR is in, lets do a formal release and test everything

[mkolodner-sc]

/unit_test

[github-actions]

GiGL Automation

@ 24:29:04UTC : 🔄 Python Unit Test started.

@ 01:40:23UTC : ✅ Workflow completed successfully.

[github-actions]

GiGL Automation

@ 24:29:04UTC : 🔄 Scala Unit Test started.

@ 24:36:25UTC : ✅ Workflow completed successfully.