michael-f-bryan / const-arrayvec Goto Github PK

I was sure this was an unsoundness bug but right before I posted I realized it's actually not, but still needs changing IMHO.

try_insert() checks if the index is bigger then the len index > self.len(), otherwise panics.
https://github.com/Michael-F-Bryan/const-arrayvec/blob/master/src/lib.rs#L241:L270

Right now it also has a check if the array full, which in my opinion should be removed (you should still be able to insert something on index 2 in a [T; 8] array even if it's full).
BUT removing that check will right now introduce an unsoundness bug, because then you could pass index = len and it will write out of bound.

the check should be index >= self.len() because doing [1u8; 8][8] is obviously out of bound and should panic per rust's semantics.

Is there some reason why <T, const N: usize> might be preferable to <[T; N]>? If const generics worked, then anything doable with the former should already be doable with [T; N], no? And [T; N] works out more ergonomic normally, no?

I like the distinction between try_insert() and insert(). Along those lines, would it make sense for the current push() to become try_push(), and implement push() something like the below?

pub fn push(&mut self, item: T) {
    match self.try_push(item) {
        Ok(_) => {},
        Err(e) => panic!("Push failed: {}", e),
    }
}