Aurelia Molzer

> But I can see why you have the huge DynamicImage enum to account for all the possibilities between GPU formats. A more comprehensive engine will certainly need all that....

PS: You might want to have a look at , I can understand the flexibility of paper (and really prefer it for sketches as well) but it's a lot clearer...

The main *future* use case for the buffer is as a separate decoding and encoding target, such that `CMYK` and `YUV` inputs can be decoded with a delayed conversion. This...

I chose not to replace `DynamicImage` right away because of compatibility, i.e. we obviously can't change the public enum to an opaque struct. Having a compatibility type sounds appealing though—a...

It's critical to understand, bgra it's not gone forever and it was removed _from the enum_ specifically—not the IO layer, so decoders and encoders may return/conume that channel arrangement. You...

Most of this is spent in `NeuQuant::init`—actually 94% of the time on my machine—called [here](https://github.com/image-rs/image-gif/blob/9206fc40daffa09183975bc0f4a257a0cfcc89a1/src/common.rs#L220-L221).

By design we can not catch double-panics which instantly abort, nor protect against malicious resource usage, so this would be seriously misnamed. I'm 100% on board with adding this as...

We should fix `png` to allow reconfiguring of limits after construction. We shouldn't require more than the default amount for intents of reading the header chunks, so this should be...

Re: `ToImageUnsafe`, what's unsafe about this? It doesn't name any invariants that the caller must uphold. Merely accessing `unsafe` methods internally doesn't necessitate something being unsafe itself. As long as...

We can't provide a zero-copy path through the encoder in any case. Most decoders don't support the on-the-fly transpose and there is no interface to configure it. That's the reason...