Thread safety

All top-level functions (load, loads, load_all, loads_all, iter_load_all, iter_loads_all, dump, dumps, dump_all, dumps_all) are safe to call concurrently from multiple Python threads. Each call owns its parser, builder, and emitter state — there is no shared mutable global.

A single Schema instance can be shared read-only across threads as long as you do not call add_loader / add_dumper on it while other threads are in flight. The registration API is not itself internally synchronised.

Individual YamlMapping / YamlSequence / YamlScalar instances follow the same rule as plain Python dict / list: they are not protected against concurrent mutation from multiple threads. Either scope one instance per thread or serialise access externally.

The tests/test_threading.py suite exercises 30 concurrent load/dump scenarios under the GIL and on free-threaded Python.