Aurelia Molzer

I wouldn't worry about the transformation into rgb in `image`. The `png` crate returns the raw color components and `image`'s `DynamicImage` is about as wrong as it is currently when...

> What to do with color-matrix values other than RGB (0 iirc). And to answer that concretely: I'd return it in an enum and _maybe_ add methods to resolve those...

The comparison to `image-webp` and `tiff` is interesting, there's lots of overlap in the required seek functionality. In particular, I can well see the use-case in building a list of...

I agree with @awxkee, this might mostly be a case of bad defaults. If we're reading large chunks of data then the overhead of a single `dyn Read` (if deeply...

Better abstractions on top of sans-IO that can communicate with an underlying transformer interface would be really neat. That way only that underlying interface needs to demonstrably preserve correctness and...

Implementation sketch: [https://github.com/197g/not-io/blob/sans-io/sans-io/tests/from_async.rs](https://github.com/197g/not-io/blob/54cbb569ad36a90156d271a450c1282de1558bd2/sans-io/tests/from_async.rs) Currently, handling of requests are done by inspecting a `Demand` enumeration that is returned from each poll of the underlying coroutine. An effect context would codify which...

Additional design nugget I stumbled onto: When we are constructing a coroutine or closure, we may _move_ some data into that type. This data is then accessibly only within the...

Soo ... with even more complication my implementation now passes under Miri. It'd would so much easier if we could simply access the coroutine state by an attribute. The hacks...

One problem with creating our own `BufReader` is that all `MaybeUninit` based reads are unstable. Hence, such a reader will necessarily have to initialize its bytes before passing them to...

> But is there a way we can get the performance we want for our more widely used decoders, avoid code size/compile time bloat, and still avoid the effort and...