Comment by phendrenad2

4 years ago

For those saying to use posix_spawn: What am I supposed to make of the writeup in the posix_spawn manpage though?

"...specified by POSIX to provide a standardized method of creating new processes on machines that lack the capability to support the fork(2) system call. These machines are generally small, embedded systems lacking MMU support"

Is this why no one uses it? It has this gratuitous opinion piece at the beginning that makes people think it's just for embedded systems and my dad's Amiga?

2 comments

phendrenad2

Reply
xenadu02  4 years ago

That's just some injected opinion, I assume from someone contributing to glibc who doesn't like posix_spawn I guess? In any case it is wrong.

Don't assume what is written in man pages is the truth. Some of them have a lot of opinion added. It can be useful to cross-check man pages between systems - they don't always call out non-portable options or behavior.

On some kernels posix_spawn is a syscall or specifies flags that make it more efficient than fork+exec. Darwin is one such system, though you can use POSIX_SPAWN_SETEXEC if you still want to replace the current process with a new executable rather than creating a child.

cryptonector  4 years ago

Hah, that's pretty funny. Regardless of the motivation as written, the motivation I surmise is:

- some systems (e.g., Windows) lack fork() for various reasons

- vfork() is baaaad

- I know, let's do something like WIN32's spawn() or CreateProcess(), but, like, better

The middle item I have good reason to think is very likely. vfork() still has a bad rap from that old "vfork() Considered Dangerous" paper. That paper circulated a lot way back when, and was the reason vfork() was removed from some Unixes for a while (well, it was left as an alias of fork()) before it was eventually re-added. The Open Group participants would been very aware of that paper, and that is almost certainly the reason that POSIX says about vfork():

  Conforming applications are recommended not
  to depend on vfork(), but to use fork() instead.
  The vfork() function may be withdrawn in a
  future version.

So if fork() can't perform well, and the committee won't recommend the use of vfork(), what shall the committee do? Answer: design and specify posix_spawn(). It's not an unreasonable answer. Though, IMO of course, they should have un-obsoleted vfork().