HN Explorer

Ask HN: Why is no one talking about kernel assisted IPC bus?

5 points by unmdplyr ↗ HN
I've been looking at kd-bus and its later evolution bus1. Seems like both projects are stalled. Why are kernel developers not supportive of a proper messaging system through kernel itself? Or rather what is stopping them from accepting patches to it?

bus1 -> https://bus1.org/

Can someone help me understand?

2 comments

[ 2.6 ms ] story [ 13.0 ms ] thread
tlb
Help me understand why Bus1 is a good idea? I'm not an expert, but it seems to me like D-bus accomplishes similar goals without any new mechanisms in the kernel.
unmdplyr
Having bus functionality enabled in kernel is beneficial for several reasons. It strips out lots of unnecessary bells and whistles and just hold some skeleton for send/receive/advertise. Most POSIX IPCs don't have an advertisement mechanism that enables discovery without some external persuasion.

D-Bus does this but has often been criticised for being overtly complex. OpenWRT came with a simpler solution, but then quickly defined a data protocol which suffers severely in I/O - it needs JSON for everything.

BUS1 doesn't specify a protocol for actual data itself so I can send/recv C struct which my process groups know about it.

Also, leaving bus logic in kernel would mean I can write a minimalistic library wrapping the syscalls for my specialised use, yet keep the applications readily portable should plans change.