> 	I was writing a program to scan ports when it started
> 	outputing something different than I expected:
> 	i.e. something goes wrong when you go above 1023.

> void main(void)
> {
> 	printf("This number %d is different from this number %d\n",
> 			ntohs(htons(1020) + htons(4)),1024);
> 	printf("This number %d is the same as this number %d\n",
> 			ntohs(htons(1020) + htons(3)),1023);
> 	printf("This number %d is the same as this number %d\n",
> 			ntohs(htons(1030) + htons(3)),1033);
> }

This is not a bug, assuming you're on a little-endian architecture,
which is presumably the case since the System: line says you're using
an i386.  (To see ntohs(htons(x)+htons(y))==x+y on a little-endian
machine, you must have (x&255)+(y&255)<=255 and (x>>8)+(y>>8)<=255, of
course for x and y each in 0..65535.  Note this is true of your "the
same as" examples and false of your "different from" example.)

You should consider network-byte-order numbers as opaque objects that
are good for nothing but (a) using in networking calls or (b) swapping
back to host order.  As you have found, on a little-endian machine,
this borders on essential; one could (and indeed I would) argue that
you should do for style and portability reasons even if you don't
expect the code to be used on little-endian machines.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu