domsson/libkita

Alternative implementation using poll or select?

Opened this issue · 3 comments

domsson commented

FreeBSD users have trouble compiling this because epoll isn't portable. Providing an alternative implementation using poll or select would solve the issue, but I have no experience with those (yet).

domsson commented

A bit of research shows that for FreeBSD, maybe kqueue might even be the better choice, as it seems to be the functional equivalent to Linux' epoll. The only issue is that I'd have to set up another machine (or virtual machine) with FreeBSD in order to develop for it.

dvogit commented

I almost forgot that FreeBSD users have some form of epoll, but is lacking in features as seen in these errors:

src/libkita.c: In function 'kita_child_kill':
src/libkita.c:1025:9: error: implicit declaration of function 'kill' [-Werror=implicit-function-declaration]
 1025 |  return kill(child->pid, SIGKILL);
      |         ^~~~
src/libkita.c: In function 'libkita_poll':
src/libkita.c:1389:2: error: implicit declaration of function 'sigemptyset' [-Werror=implicit-function-declaration]
 1389 |  sigemptyset(&sigset);
      |  ^~~~~~~~~~~
src/libkita.c:1390:2: error: implicit declaration of function 'sigaddset' [-Werror=implicit-function-declaration]
 1390 |  sigaddset(&sigset, SIGCHLD);  // default: ignore
      |  ^~~~~~~~~
cc1: all warnings being treated as errors

This comes from the package devel/libepoll-shim so I had to add /usr/local/include/libepoll-shim to the include path. I think it might be based on
an older version of epoll, but I don't really know much about Linux system calls, so make your own guesses about this.

Pulling down Linux epoll will not solve anything due to incompatible system calls. kqueue does look like a good, compatible alternative for the FreeBSD side of things.

domsson commented

Interesting! I'll see if I can get libepoll-shim to work. Two things I'd like to point out:

  • My current experiments with kqueue (which seem to mostly work) sit in the kqueue branch on the succade project, the reason being that it is easier to test there, as I can just run succade and see if things work as expected or not. Should I get kqueue to work there, then I'll copy the changes to libkita over to this repo.
  • You might notice that there is a header-only branch here at libkita. As I'm not expecting many people using this code anytime soon, I think having it as header only is probably better, which means that I'll probably merge that into master at some point. In other words, if you plan on doing some work, I'd suggest basing it off the header-only branch.

Sorry, everything is a bit of a mess. :-)