Why isn't `crypto.getRandomValues` accept a data view?

[amn]

I struggle to understand why crypto.getRandomValues doesn't accept a DataView object?

Indeed, the spec. says:

If array is not an Int8Array, Uint8Array, Uint8ClampedArray, Int16Array, Uint16Array, Int32Array, Uint32Array, BigInt64Array, or BigUint64Array, then throw a TypeMismatchError and terminate the algorithm.

Is there any reason why DataView isn't among these acceptable classes of objects? Does the procedure actually utilize different writing granularity depending on what kind of typed array it is provided, and hence it wouldn't be meaningful for it to work with a view? It could access the underlying buffer anyway, wouldn't it? I mean, this would certainly be implementation defined -- how the procedure loads the underlying buffer with the random data -- but can't a case be made that if all of the above are allowed, principally providing access to an underlying buffer, then so should be DataView? That's what it's for, isn't it? On that note, even passing ArrayBuffer could be acceptable then, on the same argument, no?

Apologies if I have missed something obvious.

[twiss]

The main reason why the spec accepts the types it does is that, when interpreting random bytes as floats, you'll get non-uniform random values, which may be unexpected. E.g., imagine someone does

crypto.getRandomValues(dataView);
dataView.getFloat32(0);

at first glance, you might expect this to produce a (uniformly) random float32.
By contrast,

crypto.getRandomValues(uint8Array);
new DataView(uint8Array.buffer).getFloat32(0);

should make it clear that we're interpreting random bytes as floats, and hopefully raise some questions about whether that's correct.
(Of course, when you're not using getFloat32/64 / Float32/64Array, then there's no issue, but the spec can't know that.)