The send/receive/reply bits are almost trivial, I can mail you the relevant parts. I am explicitly not putting any of this on GitLab or any other medium that is easily ingested by the AI peddlers, if they want my code they'll have to work for it :)

If you drop me a line on jacques@modularcompany.com I'll send you back a zip file.

In QNX, the (async) IPC is integrated with the scheduler, which is apparently responsible for the performance (still researching). So I was wondering if you chose a similar architecture. My email is "me at rochus-keller dot ch". Thanks.