On Tue Nov 25 10:52:53 2025 +0000, Rémi Bernon wrote:
I think first_window should be read with ReadNoFence at least, at it's written through atomic ops.
Isn't it a tiny bit nicer this way? The check is redundant. In the very rare potential case when it will read some stale data it will do `get_window_relative() call but will remain functionally correct with InterlockedExchange( &first_window, 0 ). But the vast majority of cases will go without fenced memory read.