hgraph 0.5.20

A functional reactive platform used to process time-series streams. Provides support for backtest (simulation) and realtime time-series processing. Using a forward propagation graph with a microtask scheduler for the runtime engine.

hgraph

A functional reactive programming engine with a Python front-end.

This provides a DSL and runtime to support the computation of results over time, featuring a graph based directed acyclic dependency graph and the concept of time-series properties. The language is function-based, and promotes composition to extend behaviour.

Here is a simple example:

from hgraph import graph, evaluate_graph, GraphConfiguration, const, debug_print

@graph
def main():
    a = const(1)
    c = a + 2
    debug_print("a + 2", c)

evaluate_graph(main, GraphConfiguration())

Results in:

[1970-01-01 00:00:00.000385][1970-01-01 00:00:00.000001] a + 2: 3

See this for more information.

Development

The project is currently configured to make use of uv for dependency management. Take a look at the website to see how best to install the tool.

Here are some useful commands:

First, create a virtual environment in the project directory:

uv venv

Then use the following command to install the project and its dependencies:

# Install the project with all dependencies
uv pip install -e .

# Install with optional dependencies
uv pip install -e ".[docs,web,notebook]"

# Install with all optional dependencies
uv pip install -e ".[docs,web,notebook,test]"

PyCharm can make use of the virtual environment created by uv to setup the project.

Run Tests

# No Coverage
python -m pytest

# Generate Coverage Report
python -m pytest --cov=hgraph --cov-report=xml

Indexing with Context7 MCP

This repository includes a baseline configuration for Context7 MCP to improve code search and retrieval quality.

• See docs/context7_indexing.md for guidance.
• The root-level context7.yaml config sets sensible include/exclude rules, priorities, and summarization hints.

Packaging, CI, and Releases

This project uses scikit-build-core and cibuildwheel to build and publish cross-platform wheels, with uv as the package/venv manager.

• Local build (PEP 517):

  # Build a wheel into dist/
  uv build -v
  # Optionally build sdist too
  uv build -v --sdist
  

• Optional Conan (developer builds only):

  # Conan is not required for packaging but may be used in local builds
  uv tool install conan
  # Example of enabling Conan provider for a local build
  CMAKE_ARGS="-DCMAKE_PROJECT_TOP_LEVEL_INCLUDES=conan_provider.cmake" uv build -v
  

• macOS notes:

  • Wheels target Apple Silicon (arm64) and require a macOS deployment target high enough for libc++ floating-point to_chars used by std::format.
  • The build system enforces MACOSX_DEPLOYMENT_TARGET=15.0 (arm64) during Release Wheels and sets a default in CMake if unset.
  • You can override locally via:

    CMAKE_ARGS="-DCMAKE_OSX_DEPLOYMENT_TARGET=15.0 -DCMAKE_OSX_ARCHITECTURES=arm64" uv build -v
    

• Windows notes:

  • Build uses MSVC (Visual Studio 2022, x64). No MinGW.
  • If stack traces via backward-cpp are enabled, Windows links Dbghelp and Imagehlp automatically.

• Linux notes:

  • Release wheels are built inside manylinux containers; optional backward-cpp integrations are disabled there for portability.

CI summary

• .github/workflows/ci.yml builds wheels on Linux/macOS/Windows for smoke validation, runs tests on Linux only (fast path), and repairs Linux wheels with auditwheel.
• .github/workflows/release-wheels.yml builds publish-grade wheels with cibuildwheel (Linux manylinux, macOS arm64, Windows AMD64), builds an sdist, validates with twine, and publishes to PyPI via Trusted Publishing.

Releasing (tag-driven)

Releases are automated via GitHub Actions with PyPI Trusted Publishing.

Prerequisites:

• Enable Trusted Publishing for this GitHub repository in the PyPI project settings (Project → Settings → Trusted Publishers). No API token is needed once OIDC is configured.
• Ensure pyproject.toml [project].version matches the tag you plan to push (e.g., 0.4.112).

Steps:

# Update version in pyproject.toml to match your release
# Then tag and push the tag to trigger the release workflow

git tag -a v_0.4.112 -m "release 0.4.112"
git push origin v_0.4.112

The workflow will:

• Build wheels for Linux (manylinux), macOS (arm64, macOS 15.0 target), and Windows (AMD64, MSVC)
• Build an sdist
• Validate with twine check
• Publish all distributions to PyPI (skips existing files)

Artifacts are uploaded for inspection as part of the workflow run.

CLion configuration tips

• Use the project venv Python for nanobind discovery:
  • CMake option: -DPython_EXECUTABLE=${PROJECT_ROOT}/.venv/bin/python
• If you need Conan, use the preamble to ensure Python/nanobind are discovered first, then the Conan provider:
  • CMake option: -DCMAKE_PROJECT_TOP_LEVEL_INCLUDES=${PROJECT_ROOT}/conan_preamble.cmake;${PROJECT_ROOT}/conan_provider.cmake
• Select the _hgraph target to build the extension.

Troubleshooting

• Nanobind not found / Python ordering in CMake:
  • Pass -DPython_EXECUTABLE=... (or -DPython3_EXECUTABLE=...) and, if using Conan, use the conan_preamble.cmake before conan_provider.cmake.
• I also find it helpful to perform an uv sync with at least --all-groups before setting up the cmake.