Programmers will want machine-precisions arithmetic (or near it -- 29 bits might work) because they feel it will be faster, use less storage, and be more likely to use "immediate" representations that don't require heap allocation.That's not really the same as "total control".

Now, the problem is that if someone wants this, the class name will be different in different implementations. So, ok, they start renaming using the module system so that they have a name they can use everywhere. That's already work that shouldn't be necessary. But they still can't write the same code for all implementations, if they want to ocnvert small ints to bignums, because some implementations require an explicit conversion. So they'll have to define a conversion function that is sometimes the identity function.

Andrew LM Shalit

The language currently handles this with the 'limited' operation, which used to create new types. You can create a subtype of integer which only includes the range of integers needed by your application. This operation is portable.

Implementations will compile that limited integer subtype to the most efficient machine representation available.