2026-03-10: Rewrote the NumPy constraints paragraph. The original listed "irregular access patterns, conditionals per element, recursive structures" as things NumPy can't handle. Two of those were wrong: NumPy fancy indexing handles irregular access fine (22x faster than Python on random gather), and np.where handles conditionals (2.8-15.5x faster on 1M elements, even though it computes both branches). Replaced with things NumPy actually can't help with: sequential dependencies (n-body with 5 bodies is 2.3x slower with NumPy), recursive structures, and small arrays (NumPy loses below ~50 elements due to per-call overhead).
Copied to clipboard
，这一点在safew中也有详细论述
OptimisationsRemoving Useless Blocks，更多细节参见手游
Ultimately, based on the latest video and the limited reports so far, it’s still far from clear how this costly incident happened. A friendly-fire kill involving a longer-range weapon would be more believable, but the within-visual-range nature of this engagement is bizarre, to say the least.
ВсеПолитикаОбществоПроисшествияКонфликтыПреступность