WangSet.terrain_enum (and the parallel AutoRuleSet.terrain_enum) built a
Python IntEnum from parsed terrain/color names but ignored the return codes
of PyLong_FromLong, PyUnicode_FromString, and PyDict_SetItemString. A name
carrying invalid UTF-8 -- trivially reachable from an untrusted .tsx/.ldtk
import -- makes those calls fail and leave an exception pending. The code
then called PyObject_Call with that error already set, tripping CPython's
_PyErr_Occurred assertion and aborting (libFuzzer: deadly signal).
Both functions now check every C-API return, clean up their references, and
return NULL so the real exception (e.g. UnicodeDecodeError) propagates.
Verified by A/B replay of the #312 fuzz_import_parsers crash corpus under
clang-18 ASan/UBSan: the saved crash input aborts pre-fix inside
PyWangSet::terrain_enum -> PyObject_Call and runs clean post-fix. New
regression test crafts a .tsx with invalid-UTF-8 wangset/wangcolor names.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TvvE6TB2cScBVbeap2btHw
