[Info-vax] Whither VMS?

[John Wallace]

On Oct 1, 6:14 pm, glen herrmannsfeldt <g... at ugcs.caltech.edu> wrote:
> Bob Eager <rd... at spamcop.net> wrote:
>
> < On Thu, 01 Oct 2009 09:07:02 -0500, Bob Koehler wrote:
> <> In article <7ij0ueF30a3u... at mid.individual.net>, Bob Eager
>
> <> <rd... at spamcop.net> writes:
>
> <>    Which makes you wonder why the inventors of C settled on such a
> <>    design in a day when CPUs were so much slower.  If I had a system
> <>    slower than a PDP-11/70, I'd have wanted it to spend it's time doing
> <>    better things.
>
> < I never saw that that, in particular, caused any kind of performance
> < problem. The libraries that used it were few - it's not often that you
> < need to know the length of a string; more often you need to know when
> < you've reached the end, and that is equally well served by a count or a
> < terminator.
>
> I have.  I have seen strcat() called inside a loop for building
> up a string while reading it in one line at a time.  The resulting
> algorithm is O(n**2).  (O(pow(n,2)) for C programmers.)  It was
> fast enough in testing, but in production the strings were millions
> of characters long and most of the time was spent in that loop.
>
> < However, they didn't want a real limit on the lengths of strings. If
> < they'd used a length prefix (as the predecessor to C did) they'd have had
> < to make it two bytes, minimum. That would have used an extra byte per
> < string, a waste of an even more valuable resource.
>
> One result of null termination is the easy buffer overflow of
> many programs today that don't properly check lengths.  Length
> at the beginning doesn't work so well if you want pointers to
> other than the beginning of a string.  The other way is with
> a structure containing a length and pointer, pretty much what
> Java does with String.
>
> < CPU time is 'extensible'; you can wait a little longer. Memory is not;
> < lack of it is a showstopper (overlays aside).
>
> Assuming that the program finishes before you need the result.
>
> -- glen

In addition to the above (thank you for saving me pointing out that
the right answer late is sometimes unacceptable ie wrong), there seems
to be a class of security exploit which involves building a string
with an embedded null followed by data which isn't ignored because the
string length processing is different depending on who's doing it.
E.g. http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2009-2666

.ASCIZ was OK for MACRO11 and similar static strings, but for dynamic
stuff, descriptors and support routines are hard to beat.