Troubleshooting

Building

CMake cannot find GMP. Install it (apt install libgmp-dev, brew install gmp); on Windows, CMake downloads it. A GMP outside the default paths is found through the GMP_DIR environment variable.

A component takes minutes where it should take seconds. Check the build type: Debug builds are 10 to 100 times slower than Release. Configure with -DCMAKE_BUILD_TYPE=Release (pip install . always builds in Release).

import wildmeshing fails from a CMake build. The module in build/bin/wildmeshing is built for the Python interpreter CMake found when it configured (Python_EXECUTABLE in build/CMakeCache.txt), and only that version imports it.

PolyFEM fails to link with access beyond end of merged section. A binutils bug with debug-info merging: add -DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=lld".

Running

An option has no effect. Check its spelling: most components ignore an option they do not know and keep its default, after a warning (qslim has no option ...). topological_offset refuses unknown options instead. wmtk_app --options <component> and help(wildmeshing.<component>) list the options; the JSON Schemas let an editor check a file before it runs.

Results change slightly from run to run. With num_threads above 0, the order of the operations depends on the schedule. num_threads: 0 runs sequentially and gives the same result every time. See Concepts: scheduling.

Simplification stops before the target. The envelope (eps_rel) refuses the remaining collapses: thicken it, or accept the larger mesh. See the QSlim tutorial.

Simplification is very slow on some inputs. Inputs where one triangle overlaps millions of small ones (some Thingi10K models) make the exact envelope test expensive. Try use_sample_envelope: true: the same envelope, tested approximately and much faster there.

TetWild returns an empty mesh. filter: flood needs a closed input; on an open one the flood fill reaches everything. Use filter: input (the input's winding number), or none. See the TetWild tutorial.

TetWild or TriWild runs to max_iterations without reaching stop_energy. Some inputs keep elements of high energy that the operations cannot improve. The stall refinement (stuck_refine_*) refines around the worst elements when the improvement stalls; the output is valid either way, with its quality in the report (max_energy, avg_energy). See Concepts: quality.

Getting help

Open an issue on GitHub with the JSON options, the input mesh if you can share it, and the log.