Delivery-Date: Mon Sep  8 20:04:31 2008
Message-Id: <200809090303.m8933c4f013124@bourbon.usc.edu>
To: cs551@merlot.usc.edu
Subject: Re: dynamic array for pid list
Date: Mon, 08 Sep 2008 20:03:38 -0700
From: Bill Cheng <william@bourbon.usc.edu>

Someone wrote:

  > Why do we need mutex for the global pid's?
  > Since the server code which will increment/decrement the counter or
  > manipulate the pid's is a single thread we dont need any mutex right?

Let's say that you use a linked list.  If you are trying to
insert a node into the linked list and this operation got
interrupted by an interrupt at a bad time, your pointer will
get messed up.  Then your code will crash later.
--
Bill Cheng // bill.cheng@usc.edu <URL:http://merlot.usc.edu/william/usc/>


On Mon, Sep 8, 2008 at 6:52 PM, Bill Cheng <william@bourbon.usc.edu> wrote:

> Someone wrote:
>
>  > I created a static array of size 50 for handling the child
>  > processes but I am having trouble managing the "int pid_list[50]"
>  > list
>  >
>  > Whenever I create a child process, I increment a global counter
>  > (and add the pid to the pid_list) in the server to reflect this
>  > and so when I break out of my "forever" loop to handle a server
>  > timeout. I try to decrement this global counter. Then I parse the
>  > list to remove the pid from this list. I use a temporary array to
>  > store the new pid_list but i get errors assigning this (address
>  > of) new list temp (smaller pid list)  as my pid list.
>  >
>  > Any other suggestions to do this? I know you mentioned the
>  > dynamic array but that seemed to require realloc() each time and
>  > I'm not sure it's a better implementation anyways.
>
> You can also use a linked list.  By the way, just because
> you have a dynamic array, it doesn't mean that you have to
> call realloc() every time.  You can increase its size, when
> needed, by 50 objects at a time.
>
>  > I was also worried about mutex for the global pid counter set in
>  > the main server program that I am modifying in a SIGCHLD handler.
>  > Should we handle this?
>
> You can use sigprocmask() to make sure that you won't get
> into a deadlock situation.  Please note that sigprocmask()
> can only be used if you use child processes.  If you use
> child threads, you should use pthread_sigmask().
> --
> Bill Cheng // bill.cheng@usc.edu <URL:http://merlot.usc.edu/william/usc/>
>
>