On Mon, Jun 23, 2025 at 2:08 PM Steffen Nurpmeso via austin-group-l at The Open Group <[email protected]> wrote: > Philip Guenther wrote in > <CAKKmsNhNfDt9cpJz-Wor5VGFrQ99=bd-s8fn_cfiescg67z...@mail.gmail.com>: > |On Monday, June 23, 2025, Steffen Nurpmeso via austin-group-l at The Open > |Group <[email protected]> wrote: ... > |> Only for the (rare) idiots among the listeners, this refers to > |> file descriptors open(2)ed etc after a fork(2), but before the > |> execve(2), which are configured with CLOFORK? > | > | It’s about all file descriptors that have FD_CLOFORK set after fork but > |before execve. > > Yes, stupid me. Maybe i meant "why did it not imply CLOEXEC".
Again, I didn't find any discussion of how this would behave with exec; I just don't think the original proposal considered it at all. To grossly over simplify, the discussions on _this_ list were mostly about the inelegance of F_DUPFD_CLO* and how that was an orthogonality warning and not really on the functionality and how it should interact with existing functionality. ... > Who has anything CLOFORK implemented as standardized already? > musl has not, glibc not in at least my variant, the BSDs do not > either. I believe MacOS has it, but no idea what version. As Alan has noted in a comment on the ticket, Solaris added support in 11.3.0 but has now amended it per my suggestion in 11.4.78 Sortix (sortix.org) implemented it in 2013 but the developer notes they never used it. Philip Guenther
