From 5c1e6b0161383b8c95a208d7a9bdfe02dd92b9ab Mon Sep 17 00:00:00 2001
From: not-matthias <matthias@codspeed.io>
Date: Sat, 16 May 2026 15:48:19 -0700
Subject: [PATCH] chore: add solution

---
 submissions/not-matthias.py | 101 ++++++++++++++++++++++++++++++++++++
 1 file changed, 101 insertions(+)
 create mode 100644 submissions/not-matthias.py

diff --git a/submissions/not-matthias.py b/submissions/not-matthias.py
new file mode 100644
index 0000000..c952f41
--- /dev/null
+++ b/submissions/not-matthias.py
@@ -0,0 +1,101 @@
+"""Parallel build scheduler for Python 3.14t (free-threading).
+
+Topo-driven worker pool. Optimizations:
+
+1. Inline chain execution — after building, if a dependent becomes ready,
+   run it directly on the same worker instead of queueing it. Collapses
+   linear chains into a tight serial loop.
+2. Batched pending decrement — workers count the length of each inlined
+   chain locally and decrement the global pending counter once per chain
+   instead of once per build. Drastically reduces global-lock acquisitions
+   on chain-heavy graphs.
+3. Main thread participates as a worker.
+4. Sharded locks for per-target remaining-deps counters.
+5. Target objects (not names) passed through the queue and dependents.
+"""
+
+from __future__ import annotations
+
+import os
+import threading
+from queue import SimpleQueue
+
+from graph import BuildGraph, Target
+
+
+_NUM_SHARDS = 64
+
+
+def build_all(graph: BuildGraph) -> dict[str, bytes]:
+    targets_map = graph.targets
+    all_targets = list(targets_map.values())
+    results: dict[str, bytes] = {}
+
+    remaining: dict[str, int] = {t.name: len(t.deps) for t in all_targets}
+    for t in all_targets:
+        t._deps_to_notify = []
+    for t in all_targets:
+        for dep in t.deps:
+            dep._deps_to_notify.append(t)
+
+    total = len(all_targets)
+    ready: SimpleQueue = SimpleQueue()
+    for t in all_targets:
+        if remaining[t.name] == 0:
+            ready.put(t)
+
+    rem_locks = [threading.Lock() for _ in range(_NUM_SHARDS)]
+    pending = [total]
+    pending_lock = threading.Lock()
+    SENTINEL: Target | None = None
+    num_workers = min(total, (os.cpu_count() or 1))
+
+    def worker() -> None:
+        local_get = ready.get
+        local_put = ready.put
+        locks = rem_locks
+        while True:
+            target = local_get()
+            if target is SENTINEL:
+                return
+            chain_len = 0
+            while target is not None:
+                deps = target.deps
+                dep_results = {d.name: results[d.name] for d in deps} if deps else {}
+                results[target.name] = target.build(dep_results)
+                chain_len += 1
+
+                next_inline: Target | None = None
+                for dep_t in target._deps_to_notify:
+                    dn = dep_t.name
+                    lock = locks[hash(dn) % _NUM_SHARDS]
+                    with lock:
+                        nv = remaining[dn] - 1
+                        remaining[dn] = nv
+                    if nv == 0:
+                        if next_inline is None:
+                            next_inline = dep_t
+                        else:
+                            local_put(dep_t)
+
+                target = next_inline
+
+            # Batch-decrement pending once per inlined chain.
+            with pending_lock:
+                pending[0] -= chain_len
+                if pending[0] == 0:
+                    for _ in range(num_workers):
+                        local_put(SENTINEL)
+
+    threads = [
+        threading.Thread(target=worker, daemon=True) for _ in range(num_workers - 1)
+    ]
+    for th in threads:
+        th.start()
+
+    worker()
+
+    for th in threads:
+        th.join()
+
+    return results