assertion failed: buckets.is_power_of_two() about hashbrown HOT 8 CLOSED

I don't understand how this assert can trigger when this one in its caller didn't.

from hashbrown.

I don't understand how this assert can trigger when this one in its caller didn't.

Maybe the compiler optimize it away (or a compiler bug)?

from hashbrown.

Here is a different trace:

from hashbrown.

I'm pretty sure this isn't an issue with hashbrown, all these cases should be unreachable. Maybe you have some memory corruption somewhere? Use after free?

from hashbrown.

To debug, I added the following code:

    fn calculate_layout_for(self, buckets: usize) -> Option<(Layout, usize)> {
        if !buckets.is_power_of_two() {
            let bits = buckets.count_ones();
            panic!("BUCKETS {} {}", buckets, bits);
        }
        
        debug_assert!(buckets.is_power_of_two());

This prints:

panicked at 'BUCKETS 8 14', /home/dietmar/.cargo/registry/src/github.com-1ecc6299db9ec823/hashbrown-0.14.3/src/raw/mod.rs:259:13

So it seems that count_ones() returns total nonsense!

Does not really look like a memory corruption, because the printed value is OK.

from hashbrown.

It works if I replace the tests with:

debug_assert!(slow_is_power_of_two(buckets));

Where slow_is_power_of_two is:

fn slow_is_power_of_two(v: usize) -> bool {
    match v {
        0b1|0b10|0b100|0b1000|0b10000|0b100000|0b1000000|0b10000000 => true,
        0b100000000| 0b1000000000| 0b10000000000| 0b100000000000 | 0b1000000000000| 0b10000000000000| 0b100000000000000| 0b1000000000000000 => true,
        _ => false,
    }
}

from hashbrown.

This seems to be a bug in rustc or in the wasm runtime you are using. You should open an issue there. If possible, try to create a minimal reproducer (using #[inline(never)] to avoid constant propagation).

from hashbrown.

OK. Many thanks for your help!

from hashbrown.