Re: RFC: Port of NetBSD cat(1)'s -f option.

On Thu, 16 May 2002, Giorgos Keramidas wrote:

> On 2002-05-16 17:40, Bruce Evans <[email protected]> wrote:
> > Using stat() instead of fstat() gives some races.  NetBSD reduces the
> > races by checking the mode both before and after open().
>
> I have the following version of basesrc/bin/cat locally:
> $NetBSD: cat.c,v 1.30 2002/05/09 02:13:10 thorpej Exp $
>
> This checks only after the open with fstat().  I agree that using
> fstat() is better.  Thanks for pointing this out ;)

Oops, I misread 1.27.  It checks only after open() too (deep in fopen()
in the cooked case).  Problems with open() are limited by using the
O_NONBLOCK flag in the -f case, but there can still be problems (open()
and/or close() can have side effects), so I think a stat() should be
tried first and checked later.

> Removed after a comment by Mike Makonnen.
>
> > - more verbose and bogus handling of the variable 'i'.  'i' is just the
> >   loop counter for a `for' loop that is obfuscated as a `while' loop.
>
> I can't think of some way to use `i' differently, without rewriting
> the `while' loop as a `for' loop too.  Mixing this change with the
> addition of -f seemed like wrong to me though.  The conversion to a
> `for' loop can be done in a separate change, if it's deemed necessary.
>
> What is it that makes you think the handling of `i' in the added code
> is bogus? :-/

The obfuscated `for' loop wasn't too bad when `i' was incremented in one
place, but `i' is now incremented in 3 places and commented on in two
places.  I first noticed the style bug of duplicating the comment.  One
comment about a simple increment may be justified because the loop is
obfuscated, but not two.

Bruce


To Unsubscribe: send mail to [email protected]
with "unsubscribe freebsd-audit" in the body of the message