Cool. So many languages get it wrong it doesn't even occur to me to check any more when learning a new language. I just assume that a mod b, where b is positive, can produce a value less than 0, and normalize the result to [0, b).

"So why doesn't C do it this way? Probably the hardware didn't do this at the time C was designed."

As far as I know, hardware still doesn't do this, if we look at the assembly instructions of CPU's. The languages that specify it towards zero have the advantage of being able to map the computation to the single CPU instruction.