🚀 go-pugleaf

In article <aeb.809273431@news.cwi.nl>, Andries Brouwer <aeb@cwi.nl> wrote:
>colin@nyx10.cs.du.edu (Colin Plumb) writes:
>: I was coming up with a list of reasons why mandating more than 64 bits
>: for now is unnecessary when I thought of a very important 128-bit
>: application: IPv6 addresses.
>
> I know a very important 4096-bit application: boot sectors.
>
> (But I don't use arithmetic on them, but instead treat them as
>   typedef struct { char a[512]; } bootsector;
> Do you use arithmetic on IPv6 addresses?)

Well, various hashing algorithms might.  And you want to do a variety
of bit-masking and comparing on them.  I agree that the manipulations
one performs on them are limited, but the fact that current 32-bit IPv4
addresses can be treated as an integral type and have things done to
them atomically is very nice.  That would be lost.

> There is no need to have integral types that are required to have
> more than 64 bits.

I tend to agree.  I just thought the IPv6 issue raised some points.
Basically, addresses count.  Somewhat inefficiently, but an address is
meant to correspond to an object (as opposed to, say, a continuous
variable where you could question how accurate the measurement is and
whether carrying more significant bits is warranted), just like file
offsets, so there's at least one example of a count larger than 64 bits
in serious use in the world.

I just hadn't thought of it, and thought others might not have, either.
-- 
	-Colin