Re: [Monotone-devel] pointer arithmetics in C++

[Zack Weinberg]

On 9/15/06, Nathaniel Smith <address@hidden> wrote:
> On Fri, Sep 15, 2006 at 10:17:05PM +0200, Markus Schiltknecht wrote:
> > I know, I know, it's evil. I want to do some pointer arithmetics
> > anyway... I need to build a hash or digest or something from three
> > 'unsigned int' values and a pointer.
>
> Is it really the pointer that you need to hash, not the value it
> points to?  Could you give more details about what exactly you're
> trying to do? :-)

It should be mentioned that pointer hashing makes your program
nondeterministic, dependent on details of memory layout - you may well
see bugs that depend on things like how much memory is required to
store all the environment variables.  As such it is almost always a
bad idea...

>   x1 ^ x2 ^ x3 ^ reinterpret_cast<int>(my_pointer);
>
> (Or perhaps multiplication would be better, or something, hash
> functions are not that easy to design.)

Note also that pointers tend to have lots of zero bits in predictable
places; I don;t know whether that makes XOR good or bad, but it's
something to think about.

> Except that this assumes that pointers and integers are the same width
> on your machine, i.e., it's not 64-bit safe, which is Bad.

size_t is the size of a pointer in all sane ABIs (i.e. everything
except weird embedded chips and 64-bit Windows); if you have C99's
inttypes.h and/or stdint.h, intptr_t there is guaranteed to be the
size of a pointer, although it may not exist.

zw