Re: [patch] sys_epoll 0.14 ...

[Andrew Morton]

Davide Libenzi wrote:
> 
> On Tue, 29 Oct 2002, Andrew Morton wrote:
> 
> > I was referring to these guys:
> >
> > +#define list_first(head)     (((head)->next != (head)) ? (head)->next: (struct list_head *) 0)
> > +#define list_last(head)      (((head)->prev != (head)) ? (head)->prev: (struct list_head *) 0)
> > +#define list_next(pos, head) (((pos)->next != (head)) ? (pos)->next: (struct list_head *) 0)
> > +#define list_prev(pos, head) (((pos)->prev != (head)) ? (pos)->prev: (struct list_head *) 0)
> >
> > if we are to add such things to list.h then lots of people need
> > to hum and hah over them first and ask questions like "why doesn't
> > it use list_empty?"  ;)
> >
> > It would be better to recode epoll's list walks to use the existing
> > list accessors.
> 
> Andrew, don't they better describe what you're actually doing instead of
> the list_empty() trick ?
> 

They are a reasonable addition to the list library.  They
should be implemented as:

/*
 * kernel-doc description goes here
 */
static inline struct list_head *list_first(struct list_head *list)
{
	if (list_empty(list))
		return NULL;
	return list->next;
}

But it shouldn't be quietly snuck in as part of epoll.   Everyone in
the world uses list.h.

Given that they are used in just a handful of places in epoll and nowhere
else in the kernel it is a little hard to justify adding them.

Unless people leap out and say "I've always wanted one of them" it would
be best to redo epoll to use

	while (!list_empty(list)) {
		item = list_entry(list, ...);
		list_del(item->list);
		...
	}

or one of the other eighty-seven list helpers which we already have.