Re: Buffer overflow Article - CACM

[tmoran@acm.org]

>In one paragraph, they criticize C as being vulnerable to such attacks
>and then dismiss Pascal as being unable to address low-level issues.
  They also say "the performance cost of bounds checking (reported in [the
'Cyclone' variant of C]) involves up to an additional 100% overhead."
I tried
      -- Lo, Hi, and A are procedure parameters, so their values
      -- and bounds are not known at compile time.
      for i in Lo .. Hi loop
        A(i) := 0;
      end loop;
with Gnat 3.15p with bounds checking on or off, -O2, and got a 65%
degradation, (Because the bounds are pushing the index out of a register?)
In the real world, my impression is that 10-15% is a more common cost of
all checking on vs all off.  Even at 65%, if the 20% of the code that
takes 80% of the time were hand checked and then compiled with checking
suppressed, 65% would change to 13% or about 3 months of CPU age by
Moore's law.

>This kind of article appears every now and then. The authors of these
>articles write as if it is necessary to improve C or invent new tools
>when all they really have to do is discover Ada.
  Sometimes one does feel like an observer watching a primitive tribe
do something the very hard way.  #.#

  They also quite fail to mention the x86's Bound instruction (does
anybody use it?) or descriptor architectures like the Burroughs machines
of 40 years ago.