Leonida's direct xgameruntime test verifies the queue and async service by looking up functions in the shim registry. The tiny game separately checks that a mapped executable can import Xbox runtime functions. A callback adds a round trip between those two boundaries. The mapped program submits work, the host calls back into it, and the callback asks the host for the operation's result. Direct test · Existing guest
We prototyped one synthetic guest for that sequence. The ordinary path passed on Wine 10. We then changed only the guest's result-query import to select a failing stub. The direct test stayed green, the guest still received its callback, and the guest correctly reported that it received the wrong result. Suppressing manual completion dispatch produced a different observation: one work call, zero completion calls.
@ItsNotRexed @IAmNotSiob, when a direct service test can pass while a loaded guest observes the wrong behavior, would you keep a small guest regression test for that boundary? This async case is our first proposal. The proposed check requires the mapped worker to run once off the caller thread, the completion to run once on the dispatching thread, and the guest to observe the worker's chosen result. It uses synthetic inputs, exercises the existing service, and can run before a title executable is available. A local implementation is ready for review.
Related work includes the original async shim and the offline services follow-up. AI-assisted implementation and testing with Codex.
For review, can someone use the failing fixture’s work count, completion count, thread IDs and observed result to decide which stage to inspect? Feedback on that diagnostic is especially welcome.
Activity
0No activity yet.