Re: [PATCH 02/18] flag parameters: socket and socketpair

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 5 May 2008 19:13:14 -0700 (PDT) Davide Libenzi <davidel@xmailserver.org> wrote:

> On Mon, 5 May 2008, Andrew Morton wrote:
> 
> > On Sun, 4 May 2008 23:42:46 -0400 Ulrich Drepper <drepper@redhat.com> wrote:
> > 
> > > +static const struct flags_rmap sock_file_flags_remap[] = {
> > > +	{ SOCK_CLOEXEC, O_CLOEXEC },
> > > +};
> > > +
> > >  asmlinkage long sys_socket(int family, int type, int protocol)
> > >  {
> > >  	int retval;
> > >  	struct socket *sock;
> > > +	int fflags;
> > > +
> > > +	type = flags_remap(sock_file_flags_remap,
> > > +			   ARRAY_SIZE(sock_file_flags_remap),
> > > +			   type, &fflags);
> > 
> > oh, so that's what it does.
> > 
> > afaict we only ever remap one or two flags.  Would it not be more efficient
> > to do that inline?  Something along the lines of:
> > 
> > 	type = flag_remap(type, &fflags, SOCK_whatever, O_whatever);
> > 	type = flag_remap(type, &fflags, SOCK_whatever2, O_whatever2);
> > 
> > ?
> 
> I think it is better to keep it able to remap more than one flag, with a 
> single call, that is table driven.
> 

The table-driven approach is slower for one and probably two bits and it is
somewhat less readable.  What do we gain in return for this?