🚀 go-pugleaf

>But an int must still accommodate the result of subtracting two
>pointers into the same array,

No, the (d)pANS has also nuked the idea that the difference between two
pointers must fit into an "int".  It must fit into *some* integral type,
and the implementation must supply an include file <stddef.h> that,
among other things, must define a "typedef" named "ptrdiff_t" that is
the integral type in question.

>so, if you're going to permit arrays of more than 2GB, you'll need
>wider ints.

No, you need wider "long"s at most.  It's not clear from a quick reading
of the standard whether an implementation that supports "long long int",
say, and that typedefs "ptrdiff_t" as "long long int", would be
conforming.  A conforming implementation implementation can, as I read
it, support "long long int", since I don't see any way for that
extension to cause a strictly conforming program to become invalid.  In
any case, it's quite possible that an implementation with 64-bit
pointers may well have 64-bit "long"s, and thus could have 32-bit "int"s.