Integer division gives different result for / and %/%

General
1

When I run the following little Stan program

data {}

transformed data {
  int NP = 22;
  int d1 = (NP*(NP-1)) %/% 2;
  int d2 = NP*(NP-1) %/% 2;
  int d3 = (NP*(NP-1)) / 2;
  int d4 = NP*(NP-1) / 2;
}

parameters {}

model {
  print([d1,d2,d3,d4]);
}

I get the following printed

[231,220,231,231]

I wold have expected the same result for all four cases, or the same result for d2 and d4. But I’m happy t be told that this result is correct (it’s easy enough to circumvent).

I’m using cmdstan v2.33.1 and cmdstanr. Code to run:

mod$sample(fixed_param = TRUE,
           iter_warmup = 0,
           iter_sampling = 1,
           chains = 1)
1 Like
2

This is the documented behavior: integer division takes precedence in the order of operations over both multiplication and division.

3

Huh, while it does appear to be intentional this does seem like it would violate the principle of least surprise.

3 Likes