[9fans] waitfor() call question?

[9fans] waitfor() call question?

[ISHWAR RATTAN]

Is it my imagination or there was a call
    Waitmesg *waitfor(int pid);

in Plan 9? I can find it in Plan9ports but not
in Plan 9.

-ishwar

Re: [9fans] waitfor() call question?

[Russ Cox]

> Is it my imagination or there was a call
>     Waitmesg *waitfor(int pid);
>
> in Plan 9? I can find it in Plan9ports but not
> in Plan 9.

There was never a waitfor on Plan 9.
There is a waitpid, or perhaps (as you suggest)
you are thinking of the one I added for use on
Unix, where it is occasionally unavoidable.

Russ

Re: [9fans] waitfor() call question?

[Joel “chesky” Salomon]

> There was never a waitfor on Plan 9.
> There is a waitpid, or perhaps (as you suggest)
> you are thinking of the one I added for use on
> Unix, where it is occasionally unavoidable.

Nemo’s book (http://lsub.org/who/nemo/9.intro.pdf, page 91) gives code
for waiting for a particular descendant process if that’s what’s
wanted.  Actually, he uses it for the immediate child process, which
makes me wonder—after the child process has exited, can the parent
wait() for a grandchild?

Just for fun:
	Waitmsg *
	waitfor(int pid)
	{
		Waitmsg *m;
		while(m = wait()){
			if(m->pid != pid)
				free(m);
			else
				return m;
		}
		return nil;
	}
To better mimic lunix semantics, a real implementation probably ought
to remember all the other pids and return them when they’re asked
for—or rather would directly try to read process information from
/proc.  (Reading the FreeBSD man page for wait(2)
(http://tinyurl.com/yy6tp2) is an educational experience in why Plan 9
is necessary ☺)

Why is waitfor() “occasionally unavoidable” on UNIX?

--Joel

Re: [9fans] waitfor() call question?

[Russ Cox]

> Nemo's book (http://lsub.org/who/nemo/9.intro.pdf, page 91) gives code
> for waiting for a particular descendant process if that's what's
> wanted.  Actually, he uses it for the immediate child process, which
> makes me wonder—after the child process has exited, can the parent
> wait() for a grandchild?

Nope.

> Why is waitfor() "occasionally unavoidable" on UNIX?

Because sometimes what would be multiple processes on Plan 9
is a single process on Unix, depending on the implementation of
the thread library, and two shared-memory procs each waiting for
their own children need not to get the other's child.
You could have some shared list of accidentally-waited-for children
but then the coordination to wake up the guy who is actually
waiting gets cumbersome.

Russ

Re: [9fans] waitfor() call question?

[Bruce Ellis]

the culprit, not the scene of the crime.

brucee

On 10/13/06, Russ Cox <rsc@swtch.com> wrote:
> > Nemo's book (http://lsub.org/who/nemo/9.intro.pdf, page 91) gives code
> > for waiting for a particular descendant process if that's what's
> > wanted.  Actually, he uses it for the immediate child process, which
> > makes me wonder—after the child process has exited, can the parent
> > wait() for a grandchild?
>
> Nope.
>
> > Why is waitfor() "occasionally unavoidable" on UNIX?
>
> Because sometimes what would be multiple processes on Plan 9
> is a single process on Unix, depending on the implementation of
> the thread library, and two shared-memory procs each waiting for
> their own children need not to get the other's child.
> You could have some shared list of accidentally-waited-for children
> but then the coordination to wake up the guy who is actually
> waiting gets cumbersome.
>
> Russ
>