Fix subtraction overflow in DuplicatesBy size_hint

[arpankapoor]

Sample that panics with attempt to subtract with overflow in a debug build:

use itertools::Itertools;

fn main() {
    println!("{:?}", vec!["a", "b", "c", "c"].iter().duplicates().collect_vec());
}

[phimuemue]

Thanks for this. Yes, cmp::max(a-b, 0) seems strange on usize. I do not have much time right now, so just two questions:

• Do you know if there are other occurences of this pattern in the code? (A "no, I do not know" is fine, we should just look ourselves, then.)
• Should we add tests for this case?

[arpankapoor]

• I haven't seen this pattern repeated elsewhere on a cursory glance.
• Added a test similar to the above example. Let me know if that seems sufficient.

I've also changed the checked_sub call to saturating_sub as per clippy's suggestion.

[phimuemue]

Hi there, one thing that bothers me is that cmp::min and saturating_sub actually answer the same question twice. (Namely: is hi <= self.meta.pending?)

I think it would be clearer to do this:

if hi <= self.meta.pending {
    // fewer or equally many iter-remaining elements than pending elements
    // => at most, each iter-remaining element is matched
    hi
} else {
    // fewer pending elements than iter-remaining elements
    // => at most:
    //    * each pending element is matched
    //    * the other iter-remaining elements come in pairs
    self.meta.pending + (hi-self.meta.pending)/2
}

[arpankapoor]

updated as per your suggestion

[phimuemue]

Thank you.

bors r+

[phimuemue]

Merged as this improves the current size_hint. Maybe we should use our size_hint::add et al instead of plain + and others.

[bors]

Build succeeded:

• bors build finished