This covers the first part of the rules but will quickly reach overflow if you combine operations without using the second rules to reduce scale:

  private static Decimal adjustScaleIfNeeded(int precision, int scale) {
    if (precision > MAX_PRECISION) {
      int minScale = Math.min(scale, MIN_ADJUSTED_SCALE);
      int delta = precision - MAX_PRECISION;
      precision = MAX_PRECISION;
      scale = Math.max(scale - delta, minScale);
    }
    return new Decimal(precision, scale, 128);
  }

@westonpace I'm having a hard time following the logic behind adjustScaleIfNeeded. I can understand truncating the scale of the output in order to make it fit, however, the MIN_ADJUSTED_SCALE part doesn't make sense to me. It acts to clamp the scale to be <= 6, but without this having an impact on the output precision, which seems to be incorrect?

Say the output had a scale of 10 and a precision of 40, adjustScaleIfNeeded would then clamp this to a scale of 6 and a precision of 38. I would have thought it would either clamp to a scale of 8 and precision of 38, or a scale of 6 and a precision of 36?

Edit: oh no nvm, brain fart. It exists for the case of a scale of 5 and a precision of 40, it will refuse to reduce the scale further. This does mean that such an operation could still overflow though.

I've filed #4664 and have a preliminary implementation of what it would take to support precision-loss decimal arithmetic. I hope to polish this up over the coming week

Decided to not move forward with this, see #4664 (comment)