In order to have sockets on UNIX domain, do you think this could help ?
http://lxr.linux.no/source/net/unix/sock.c?v=1.0.9#L669
(i tried to reach the mailing list, but the sign-up link doesn’t work)
Quote:
Well, I'm not fluent in Linux UNIX domain sockets implementation, but basicly their approch seems to have for each AF_UNIX socket a buffer that they shared all AF_UNIX domain sockets. When a socket write, the data are just append/written to the AF_UNIX socket peer's buffer.
In order to have sockets on UNIX domain, do you think this could help ? http://lxr.linux.no/source/net/unix/sock.c?v=1.0.9#L669
That’s a way to do it, indeed.
I just find it “bloaty”.
xBSD implementation use the VFS inode own buffers to do it, though, which sounds even more complex and bloated to me.
I did search a little but failed to find a quick & easy implementation using for example named pipes. We do have a working pipefs in Haiku, I can’t see why we should reimplement buffering but, well, dunno.
Quote:
Try subscribing from this page then:
http://www.freelists.org/list/openbeosnetteam
(i tried to reach the mailing list, but the sign-up link doesn't work)
I’ll fix the broken link from team page, thanks for reporting…
Why not creating a named pipe by mkfifo for a socket on a file ?
It seems it could be an interesting trade-off to test some things, and maybe just keep this kind of implementation.
If not, i would be interesting on your point of view 