Set up Cursor Cloud dev environment for jupyddl

[guilyx]

Summary

Sets up a working development environment for jupyddl (a Python wrapper over the Julia PDDL.jl parser via pyjulia/PyCall.jl), documents it for future cloud agents, and repairs the broken GitHub Actions CI.

Dev environment (captured in the VM snapshot)

• Julia 1.5.2 (/opt, documented tested version) + Julia packages PyCall.jl (built against the venv Python) and the APLA-Toolbox/PDDL.jl fork.
• Python 3.8 (deadsnakes, dynamically linked — required so pyjulia can embed Julia; static uv/standalone Pythons break the bridge) in .venv with julia, coloredlogs, matplotlib, pytest, pytest-cov, flake8.
• python3.8-tk (the code picks the TkAgg backend when a display is present) and initialized the pddl-examples submodule.

Update script (registered)

git submodule update --init
test -d .venv || "$HOME/.local/bin/uv" venv --python /usr/bin/python3.8 .venv
"$HOME/.local/bin/uv" pip install --python .venv/bin/python -r requirements.txt pytest pytest-cov flake8

CI fix

The build and tests workflows failed at actions/setup-python@v3: Version 3.7 with arch x64 not found. The step hardcoded Python 3.7 (ignoring the matrix), and 3.7 has no build for the current ubuntu-latest (24.04) runner.

• Use ${{ matrix.python-version }} and pin Python 3.8 (newest interpreter whose resolved deps keep NumPy at 1.x; matplotlib==3.5.1 breaks on NumPy 2.x, which is pulled in on Python ≥3.9).
• Drop the macos-latest leg: it is now arm64, which has no Julia 1.5.2 build and no EOL-Python build, so the pinned toolchain can't run there.
• Build PyCall against the runner's Python via python -c "import julia; julia.install()" before adding PDDL.jl (with a corrected url= PackageSpec); check out submodules via actions/checkout@v4; modernize action versions to avoid deprecated-Node failures.

All three previously-failing checks are green: build (ubuntu-latest, 3.8) (both workflows) and test (ubuntu-latest, 3.8).

Testing

• Hello-world: parsed pddl-examples/dinner and ran BFS, A* (h_add / critical_path), and Greedy planners — all return plan op2(A, A, C).
• CLI: scripts/ipc.py produced a full 8-heuristic A* metrics report.
• Reproduced the CI pipeline locally (isolated Julia depot, headless): on Python 3.8 flake8 gate is 0 and pytest --cov=./ → 77 passed. Python 3.10 was rejected (NumPy 2.x ABI break), confirming the 3.8 pin.
• Verified the fixed workflows pass on GitHub Actions.

Notes

AGENTS.md documents the non-obvious bits: the dynamically-linked-Python requirement for the Julia bridge, how to rebuild PyCall if the venv path changes, the DISPLAY-dependent DataAnalyst test behavior (all 77 pass headless; 19 fail only when DISPLAY is set, due to a pre-existing unsorted-os.walk domain/problem swap), and the CI Python pinning rationale.

  

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[cursor]

Risk: medium. Approved CI workflow modernization and AGENTS.md documentation; Cursor Bugbot was not present on this PR. No reviewers assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}