serp-threading

Drop-in replacement for CPython's threading module (D9 v2). Thread and Timer are language intrinsics; the sync primitives are runtime-held handles (Mutex/Condvar tables in the runtime, no crates). Under CPython the serp_threading_core delegate mirrors the same state machines on stdlib threading, so output is byte-identical across both worlds.

API

• Thread(target=f, args=(...), name=None, daemon=False) → .start(), .join(timeout=None), .is_alive(), .name, .ident, .daemon. Timer(interval, function, args=()) adds .cancel(). The target is a named module function whose parameters are Send-by-copy: int/float/bool/str/bytes, or Owned[C] where C is a dataclass made only of those scalars (the sync primitives below qualify). Non-daemon threads are joined when main() returns, like CPython.
• Lock, RLock — acquire(blocking=True, timeout=-1), release(), locked(), with.
• Event — set/clear/is_set/wait(timeout=None).
• Semaphore(value=1), BoundedSemaphore(value=1) — acquire(blocking, timeout), release(n=1), with.
• Condition(lock=None) — acquire/release/wait(timeout)/wait_for(pred, timeout)/notify(n)/notify_all(), with.
• Barrier(parties, timeout=None) — wait(timeout), reset(), abort(), parties, n_waiting, broken; BrokenBarrierError.
• local() — thread-local store: set(name, value), get(name, default=None), has(name), delete(name).
• current_thread(), main_thread() (.name/.ident/.daemon), get_ident(), get_native_id(), active_count(), TIMEOUT_MAX.

Divergences

No shared mutable objects cross the thread boundary — hand data over through serp_queue (deep-copied PyVal items) or thread return values. Thread targets must be module functions (no lambdas/bound methods, no kwargs=); Thread.name is read-only; local() uses get/set/has/delete instead of attribute access; uncaught thread exceptions print Kind: message to stderr; no enumerate(), excepthook, settrace/setprofile, stack_size.

Install

serp add serp-threading
pip install serp-threading