Allow s6-svscan to handle more signals about s6 HOT 2 CLOSED

It's not only about looking for a handler script when a signal arrives. s6-svscan has to explicitly catch the signal first. That's why you couldn't make SIG??? work - the signals you wanted are simply not caught.

There's no major theoretical objection to what you want, but there are a few practical ones, that makes catching each and every signal very difficult to do portably. One of them is that there is an implementation-defined set of signals you just cannot trap, and on some systems sigaction() fails, and making sure the failure is due to an uncatchable signal and not a kernel issue is a problem that requires either a lot of code, or #ifdef mazes, to be solved correctly. Other issues are, for instance, how to establish a bijection between a signal such as SIGRTMIN+3 and its handler script for an (almost) arbitrarily large number of signals, without breaking compatibility with existing names, and without being entirely counter-intuitive (SIG13, SIG14, SIGTERM, SIG16, what?)

s6-svscan aims to be portable, and is a very critical piece of code, so I want to keep it as simple as possible. I had already thought about generalizing the signal diversion mechanism, but in the end the benefits aren't worth the amount of code and complexity that it would add. I'm sorry. You'll end up with much less complexity on your system by just setting lxc.signal.halt. :-)

from s6.

No worries. Thought it couldn't hurt to ask :) I figured it was something more complex than I was aware of in my limited C knowledge. I'm developing some generic LXC images and I was trying to avoid having to tell people to always set lxc.signal.halt, but sounds like that's the easiest solution. Thank you for the detailed explanation!

from s6.