Verify decoding works correctly in a subinterpreter.

The PyDict_Next fast path (unsorted exact dict) must not leak references compared to the iterator slow path.

The iterator slow path (sorted or dict subclass) must not leak.

Slow path: dict subclass with sort_keys=True.

Slow path: dict subclass falls back to iterator path.

Empty dict takes the early-return path.

Verify encoding works correctly in a subinterpreter.

Verify Encoder heap type instances encode correctly.

Slow path: exact dict, sort_keys=True.

Fast path: exact dict, sort_keys=False.

Fast path: exact list iterates by index.

Fast path: exact tuple when tuple_as_array=True.

Error path in scanner_new/encoder_new must release module_ref.

for_json() raising inside its body must not leak the method binding or partial accumulator state.

Verify _speedups can be imported in a subinterpreter.

A test against the regression mentioned at `github issue 29`_. The indent parameter should accept any type which pretends to be

Encoding a list of dicts with the same keys exercises the key_memo cache in encoder_encode_dict_key.

Large dict stresses the PyDict_Next iteration path.

Large list stresses the indexed fast path.

The indexed fast path for exact lists must not leak.

Slow path: list subclass uses iterator.