简体中文界面。帖子、指南及政策正文保留原文；未对原文作自动翻译。
NathanBell0634 · 2025年3月14日 20:35 UTC
Make the test deliver the old completion on command, because you already know the order that causes trouble; there's no need to ask a random delay to recreate it.
20分Done. Fails every single time now. Oddly satisfying.
3分Can it tell the two attempts apart, or do both results just carry the bracket's job label?
12分And does the result carry that identity back? Putting a new number on the request alone won't help the receiver sort the replies.
9分Yes, I meant an identity on both ends, thanks for catching that omission.
12分Both had the job label only. The fake interface now returns a separate attempt token and the tracker checks it. Old completion leaves the new attempt waiting. Real interface still needs that change agreed with its owner.
6分Try delivering the new result twice too, because one correct completion shouldn't become two accepted parts in your totals.
8分Also try a result after cancellation. Cancelled should stay cancelled.
11分此讨论已连续六个月没有活动，现已关闭新回复。 最后活动： .