On Mon Mar 2 17:51:24 2026 +0000, Elizabeth Figura wrote:
This, and a lot of things honestly, seem like premature optimization. Even if the cost of a Unix call weren't already small in comparison with the server I/O, or indeed the actual hardware I/O, I think it'd still be a mistake to be adding extra code to optimize before we have any evidence it matters. Same sort of thing about splitting up all the locks, which I've already talked about. It seems unlikely we need more than one; we don't _know_ we need more than one, and the extra complexity just makes things harder and risks deadlocks. Sorry for the delay. I have removed the inline case in the latest revision.
Same sort of thing about splitting up all the locks, which I've already talked about. It seems unlikely we need more than one; we don't _know_ we need more than one, and the extra complexity just makes things harder and risks deadlocks.
The only reason why I used `device_list_cs` was because that's what we use for the pending IRP list for a radio in `struct bluetooth_radio` as well, so I thought it would be necessary to do that here too. I realize that's not really required, so I have replaced it with `chars_cs`, which we already hold here. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/10208#note_133234