Replace socket2 calls with rustix #146
Conversation
| // Set nosigpipe if necessary. | ||
| #[cfg(any( | ||
| target_os = "macos", | ||
| target_os = "ios", | ||
| target_os = "tvos", | ||
| target_os = "watchos", | ||
| target_os = "freebsd" | ||
| ))] |
There was a problem hiding this comment.
AFAIK, NetBSD also supports nosigpipe (OpenBSD and DragonflyBSD do not).
- SO_NOSIGPIPE https://man.netbsd.org/setsockopt.2
- SOCK_NOSIGPIPE https://man.netbsd.org/socket.2
Also, do you know why socket2 sets nosigpipe only on Apple targets, not also on FreeBSD/NetBSD?
https://github.com/rust-lang/socket2/blob/66ed6b055a4352a26ed25e8d01981fb5b555f3d8/src/socket.rs#L798
Maybe because MSG_NOSIGNAL has already been passed?
https://github.com/rust-lang/rust/blob/fece303aeb1d2b845745bc3c067475619670a868/library/std/src/sys_common/net.rs#L38
There was a problem hiding this comment.
I don't know, I would bet that it's because FreeBSD/NetBSD did not have it at the time.
There was a problem hiding this comment.
Opened bytecodealliance/rustix#870 for NetBSD and DragonFly BSD.
I think we can set HANDLE_FLAG_INHERIT by using SetHandleInformation from windows-sys for now. (We already depend on windows-sys via polling and rustix.) Ideally, I would like to set that at the time the socket is created (i.e., by using SocketFlags). |
socket2 is our last libc-using dependency. By replacing it with rustix calls, we can make this crate libc-free. This doesn't set the inherit disable property on Windows yet, as rustix does not support it. Signed-off-by: John Nunley <dev@notgull.net>
Signed-off-by: John Nunley <dev@notgull.net>
Signed-off-by: John Nunley <dev@notgull.net>
notgull
force-pushed
the
notgull/no-socket2
branch
from
1855eca
to
7eeecbc
Compare
Signed-off-by: John Nunley <dev@notgull.net>
Signed-off-by: John Nunley <dev@notgull.net>
Signed-off-by: John Nunley <dev@notgull.net>
|
The OpenBSD test failure is rust-lang/rust#116523 combined with bytecodealliance/rustix#869 |
Signed-off-by: John Nunley <dev@notgull.net>
|
I've worked around this issue by disabling |
In #146, I introduced a bug that prevented abstract sockets from working. I passed the path straight into rustix::net::SocketAddrUnix::new, which fails if it receives an abstract socket. This commit fixes this issue by explicitly checking for abstract sockets. If it sees that the path it's receiving is abstract, it will pass the path's bytes to new_abstract_socket() instead. This should fix the issue that is occurring in dbus2/zbus#517 Signed-off-by: John Nunley <dev@notgull.net>
In #146, I introduced a bug that prevented abstract sockets from working. I passed the path straight into rustix::net::SocketAddrUnix::new, which fails if it receives an abstract socket. This commit fixes this issue by explicitly checking for abstract sockets. If it sees that the path it's receiving is abstract, it will pass the path's bytes to new_abstract_socket() instead. This should fix the issue that is occurring in dbus2/zbus#517 Signed-off-by: John Nunley <dev@notgull.net>
socket2 is our last libc-using dependency. By replacing it with rustix calls, we can make this crate libc-free.
This doesn't set the inherit disable property on Windows yet, as rustix does not support it.