feat(python): packaging python projects and uploading to pypi

[yangxk1]

Describe the enhancement requested

As shown in the title.

We registered the pypi name of graphar.

It is also necessary to

• adjust the python project structure
• improve pyproject.toml
• create CI workflow files

Component(s)

Python

[yangxk1]

#813 provides an initial solution for automatically building Python wheels and publishing them to PyPI. The current build targets macOS and Linux.

Trigger conditions:

• When triggered by a PR, the workflow builds the packages but does not upload them to PyPI.
• On every commit pushed to main, the workflow uploads to TestPyPI only (it does not upload to PyPI).
• For an official release, the workflow can be triggered manually and will upload to PyPI.

Versioning:

• The version is updated automatically via .github/scripts/update_version.py.
• Only the TestPyPI version is updated to avoid version conflicts.
• For releases to PyPI, the version depends solely on the setting in python/pyproject.toml.

Possible improvements:

• A more appropriate versioning strategy.
• More suitable triggering cadence—for example, switching to daily/weekly runs when PR volume increases.
• Add support for Windows x86_64.

[yangxk1]

We can now install GAR using pip install graphar. There is a development version 0.13.0.dev0 in pypi. The first semantic version will be released together with 0.14.0.

cc/ @adsharma

[adsharma]

@yangxk1 got a missing symbol error on arrow::fs::InitializeS3(arrow::fs::S3GlobalOptions const&)

$ uv venv
$ source .venv/bin/activate
$ uv pip install graphar

(incubator-graphar-testing) ➜  incubator-graphar-testing git:(main) ✗ cat test.py
import graphar

# Load graph info from a YAML file
graph_info = graphar.graph_info.GraphInfo.load("nebula/basketballplayergraph.graph.yml")

# Access vertex information
vertex_info = graph_info.get_vertex_info("person")
print(f"Vertex type: {vertex_info.get_type()}")

# Access edge information
edge_info = graph_info.get_edge_info("person", "knows", "person")
print(f"Edge type: {edge_info.get_edge_type()}")
(incubator-graphar-testing) ➜  incubator-graphar-testing git:(main) ✗ uv run test.py
Traceback (most recent call last):
  File "/home/ubuntu/src/graphar/incubator-graphar-testing/test.py", line 1, in <module>
    import graphar
  File "/home/ubuntu/src/graphar/incubator-graphar-testing/.venv/lib/python3.13/site-packages/graphar/__init__.py", line 20, in <module>
    from .graph_info import *
  File "/home/ubuntu/src/graphar/incubator-graphar-testing/.venv/lib/python3.13/site-packages/graphar/graph_info.py", line 18, in <module>
    from ._core import DataType, Property, PropertyGroup, AdjacentList, VertexInfo, EdgeInfo, GraphInfo
ImportError: /home/ubuntu/src/graphar/incubator-graphar-testing/.venv/lib/python3.13/site-packages/graphar/libgraphar.so: undefined symbol: _ZN5arrow2fs12InitializeS3ERKNS0_15S3GlobalOptionsE

[adsharma]

Potential workaround:

The issue was that graphar's bundled libgraphar.so was built with S3 support but the bundled libarrow.so in graphar.libs/ didn't have the required S3 symbols. Manually replacing it with the version from pyarrow with S3 support works.