Changelog for v0.24

[HeroicKatora]

Closes: #1433

[fintelia]

Looks good to me. Pointed out some typos.

[HeroicKatora]

Upgrading quickcheck made it discover some inconsistencies in our resize_dimensions code. I'm checking those before publishing.

[val-ms]

Does this mean we'll get a new release soon? 😃

[val-ms]

Upgrading quickcheck made it discover some inconsistencies in our resize_dimensions code. I'm checking those before publishing.

@HeroicKatora I don't know if this is related to resize_dimensions, but it appears that resize_exact gives me more accurate resizes in the unreleased master branch than in 0.23.14. I'm attributing the improvement to it doing the calculations internally using an ImageBuffer<Rgba<f32>, Vec<...>> type.

For context: I'm trying to replicate fuzzy hashes using your image crate as the Python imagehash::phash() function. My Rust implementation is off by a (literal) bit on some hashes when using image v0.23.14 but I get exact matches for my current sample set if I use the master branch.

[HeroicKatora]

Time for some findings. It looks like the quickcheck based test diverged from the implementation in that the special case of ensuring a non-empty result was never introduced. Somehow, the previous version (quickcheck = 0.9) also did not find this case. After that change, we find some cases in very large numbers:

thread 'math::utils::test::resize_bounds_correctly_width' panicked at '[quickcheck] TEST FAILED. Arguments: (1, 10737419)', […]/src/github.jparrowsec.cn-1ecc6299db9ec823/quickcheck-1.0.3/src/tester.rs:165:28

This case should resize one dimension to u32::MAX.

Diff for the fix to the implementation

diff --git a/src/math/utils.rs b/src/math/utils.rs
index 1ed129cf..1999f350 100644
--- a/src/math/utils.rs
+++ b/src/math/utils.rs
@@ -45,7 +45,9 @@ mod test {
         fn resize_bounds_correctly_width(old_w: u32, new_w: u32) -> bool {
             if old_w == 0 || new_w == 0 { return true; }
             let result = super::resize_dimensions(old_w, 400, new_w, ::std::u32::MAX, false);
-            result.0 == new_w && result.1 == (400 as f64 * new_w as f64 / old_w as f64).round() as u32
+            let exact = (400 as f64 * new_w as f64 / old_w as f64).round() as u32;
+            println!("{:?}, {:?}", result, (new_w, exact.max(1)));
+            result.0 == new_w && result.1 == exact.max(1)
         }
     }
 
@@ -53,7 +55,9 @@ mod test {
         fn resize_bounds_correctly_height(old_h: u32, new_h: u32) -> bool {
             if old_h == 0 || new_h == 0 { return true; }
             let result = super::resize_dimensions(400, old_h, ::std::u32::MAX, new_h, false);
-            result.1 == new_h && result.0 == (400 as f64 * new_h as f64 / old_h as f64).round() as u32
+            let exact = (400 as f64 * new_h as f64 / old_h as f64).round() as u32;
+            println!("{:?}, {:?}", result, (exact.max(1), new_h));
+            result.1 == new_h && result.0 == exact.max(1)
         }
     }

[fintelia]

Ideally we shouldn't be mixing code changes, dependency bumps, and the changelog updates all in the same PR. However, everything here looks good to me.