serde-this-or-that VS kache

GitHub | crates.io | Changelog

I was just working over the weekend on putting together a project, serde-this-or-that which is inspired by a blog post of a similar name, even though technically that blog post appears to discuss something slightly tangential to the matter. This project is inspired by a question posted on SO, which suggested a crate might be useful to deserialize cases where a json field has one of multiple possible multiple types (for example it could come as str or bool). I was also working on adding benchmarks, which this is actually the first crate where I have successfully put together benchmarks using criterion which I've been kind of nervous about using in general, but based on the docs it was pretty easy to set up so I'm kind of glad about that at least.

Update: for those curious, I've added benchmarks for comparison with serde_with here. After some more testing, I've release v0.4 and am confident that performance is now on par with serde_with. Also, I noticed that whenever serde_with::PickFirst is used, performance drops; in this case, serde_this_or_that does seem to perform 12x better in such scenarios.

The full release notes can be found on the project page on GitHub as well.

I've just published an initial release of serde-this-or-that, which builts on top of serde to provide custom deserialization for fields that can be specified as multiple types.

When writing https://github.com/cargo-quick/kache we found that the async compression crate was set up nicely for streaming with minimal overhead, but didn't support multithreaded compression (which the zstd crate does support in particular modes). In practice, we are targeting GitHub Actions runners, which only have 2 cores, so it didn't matter much.