On Wed Sep 10 18:11:22 2025 +0000, Matteo Bruni wrote:
Alright, I think I fixed up all of the above. I have some comments though... I don't have particularly good "versions" to put in the `broken` test comments in patch 7/10. Basically, current Windows 10 AND 11's behavior is different from older Windows 10 / 11. Also, for Windows 10 at least, we're talking about the same 22H2 (or 19045) version. There is a "revision" value accessible through the registry which can help a bit (see e.g. https://learn.microsoft.com/en-us/windows/release-health/release-information, that's the value after the dot under the Build column) Notice also that those revision numbers are specific for each different OS version and can, and will, clash among different versions. I have a small patch to print that value from winetest, since it seems to be generally useful information to have. That doesn't help a ton here though. I know that the newest Win10 revision on the testbot and test.winehq.org (19045.2311) has the "old" ntoskrnl tests behavior, while my own 19045.6216 has the "new" one. I could make the comments a bit more thorough but I'm not sure that's much of an improvement.
The other comment is about the new ntoskrnl tests. I don't understand why `CancelIoEx()` "by itself" seems to block in this version of the tests. It's quite puzzling given that effectively the same `CancelIoEx()` doesn't seem to block in the later test when it's followed by `CancelIo()` (which does block as expected).
I'm going to try some more later but there's a good chance that there's some issue with the tests or I'm missing something obvious. Any help is appreciated!