Let's try to document our roadmap. We should probably identify things are "nearer term" and "longer term / wishlist / ideas".

Here's a brainstorm (i.e., probably too much, and not well organized or described):

• Automate building wheels of python-suitesparse-graphblas for major platforms using cibuildwheel
• Improve user-define types (UDTs) and enable easier "auto-vectorization"; see Best practices for vectorising GraphBLAS #298
• Add GPU support (once SuiteSparse:GraphBLAS officially supports it)
• Keep improving documentation
  • docstrings, docs website, examples, tutorials, benchmarks, contributing, etc.
  • Open documentation issues
• Advocacy
  • blog posts, talks, tutorials, social media posts, connect with influencers, etc.
• Find more applications and partnerships for GraphBLAS in the PyData / Scientific Python ecosystem
• Keep trying to become more of a community project
  • Easier onboarding, have ambassadors, etc.
  • More roles for new members, etc.
• Improve code maintainability
  • Add type annotations
  • Improve dev documentation where appropriate (as code comments, docstrings, etc.)
• graphblas-algorithms
• Improve NetworkX dispatching
• Distributed GraphBLAS in Python with dask-graphblas
  • Finish renaming from dask-grblas and move to python-graphblas org
  • Update to run with latest python-graphblas
  • Update API to better match latest python-graphblas
  • Benchmarks!
• Support multiple GraphBLAS backends (ENH: Add "pure GrB" backend that doesn't rely on SuiteSparse-specific behavior #172)
• MLIR-GraphBLAS backend
• Pure Python GraphBLAS backend
• Expose LAGraph in Python
• https://github.com/GraphBLAS/binsparse-specification
• Wrap python-graphblas to match the Python array API standard
• Create cython-suitesparse-graphblas to allow using SuiteSparse:GraphBLAS in Cython.