Was there an issue with the original code?

BLOCK_MAXSIZE is a usize right?

I wrote this to make the tests pass, but I'm sure this could be optimized.

Do you think this would suffice?

if (<usize>size > BLOCK_MAXSIZE) throw new RangeError(E_INVALIDLENGTH);

I don't know if there is a range gap regarding negative numbers and casting to usize. We just need to make sure all the negative numbers throw an error because we accept a i32 instead of a usize.

I remember that the original idea was that a negative size would wrap around when cast to unsigned, in turn becoming larger than BLOCK_MAXSIZE (which is < 2^30) when cast to unsigned, so we can use a single instruction. I also remember that there used to be issues with that, just not sure if this line is also affected?

Right. Thank you for reminding me that we had a conversation about this.

When any negative i32 is cast to usize will it always be bigger than BLOCK_MAXSIZE? I suppose that depends on if we are running in 64 bit wasm right? As long as the max blocksize is (1 << 30) - BLOCK_OVERHEAD then the assertion appears to hold true.

Sounds like we should optimize this check away. :)

Even more, I confirmed that this works practically by spinning up an as-pect test randomly that looked like this:

  it("should test something", () => {
    Expected.report(true);
    for (let i = i32.MIN_VALUE; i < 0; i++) {
      let condition1 = (i32(<usize>i > BLOCK_MAXSIZE) | i32(i < 0));
      let condition2 = i32(<usize>i > BLOCK_MAXSIZE);
      let result = condition1 == condition2;
      Actual.report(result);
      assert(result, "Result should match.");
    }
  });

The assertion seems to hold true, practically. Thanks for the reminder.