Release v6.2.10

This is a patch release fixed a critical bug in TTLCache/VTTLCache (popitem() method), and focused on correctness, deadlock avoidance, and internal robustness.

What's Changed

• All cache methods now detach GIL when waiting for the lock. Follow-up to the #95 deadlock fix. Methods such as __len__, __bool__, clear, popitem, capacity, and the TTL expire helpers now detach GIL before locking (lock_py_attached). (#117)

• __eq__ / __ne__ lock the two caches in address order. Prevents the classic cross-thread deadlock where one thread does a == b while another does b == a`. (#122)

• update() no longer runs Python code under the cache lock. For anything other than an exact dict/list/tuple/set/frozenset, the pairs are built outside the lock in 1024-entry chunks so peak memory tracks the cache size, not the input size. (#123)

• popitem() on TTLCache/VTTLCache now drops expired entries first. It no longer returns an already-expired key. (#121)

• When a C-level key’s tp_hash or tp_richcompare fails without setting an exception, now cachebox raises SystemError instead of panicing. (#120)

• The internal linked list implementation now is more safer. linked list stays consistent if an element’s Drop panics. (#124)

• Documentation corrections. (#119)

Special Thanks to

• @greenclaw for the deadlock fixes (#117, #122, #123), the SystemError hardening (#120), the LinkedList panic-safety fix (#124), the docs corrections (#119).

Full Changelog: v6.2.9...v6.2.10