[xylar]

cull() and convert() in mpas_tools.mesh.conversion created their working directory with mkdtemp() and never removed it, so every call left copies of the input mesh, masks and output mesh behind (in the current directory when dir='.' is passed, as Polaris does). This PR uses TemporaryDirectory, as mask() already does, closing the loaded output dataset and copying any graph.info file before the directory is removed. It also adds a test that only the requested graph files remain in dir after culling and converting a planar mesh.

Fixes #764

---

Posted by Claude Code on @xylar's behalf. The testing, analysis and wording above are AI-authored; please check them accordingly.

🤖 Generated with Claude Code

[xylar]

Testing

mpas_tools 2.2.0 isn't on conda-forge yet, so I tested in a conda environment with the mpas_tools 2.1.0 package (for the compiled executables) and this branch installed in editable mode on top.

• The new test_cull_convert_remove_temp_dir fails on master: a tmpXXXXXXXX directory is left behind next to the graph files. It passes with this branch.
• From the repo root, pytest conda_package/tests/test_conversion.py conda_package/tests/test_mesh_mask.py conda_package/tests/test_sort_mesh.py gives 11 passed and 1 failed. The failure is test_conversion_triangle_angle_quality, which is expected with the 2.1.0 executables: that test needs the triangleAngleQuality output from the 2.2.0 converter (Fix law-of-cosines typo in triangleAngleQuality #763).
• No tmp* directories are left in the working directory after the test run.
• pre-commit passes.

---

Posted by Claude Code on @xylar's behalf. The testing, analysis and wording above are AI-authored; please check them accordingly.