Jon Sneyers

I agree it would be useful to document and clarify the limitations of what kind of JPEG files can be losslessly transcoded. If I recall correctly, we at some point...

My idea was to put R, avgG, B in the main image, and G1-G2 in an extra channel. Compression-wise, that should work quite well. The main challenges are: - preserving...

Could be an OOM problem, since the image is rather large. What happens if you reduce the effort, e.g. `-e 3`?

Ah, could it be that it's trying to address more than 2^31 bytes (2GB) of memory and that causes trouble on 32-bit platforms? If that's the issue, there's not much...

To clarify, the current encoder takes a whole image as input and encodes it with global optimization of the entropy coding etc. In this case, the image itself is 38.4...

I think this might be tricky to do in a portable way...

General-case modular jxl decoding is expected to be slower than png or webp decoding. One thing that might be different though is how multi-threading is being used. Perhaps the plugin...

I propose to do both: - implement progressive alpha in chrome (it should already work in current libjxl, afaik) - also strip opaque alpha by default — too many things...

Presumably the jxl was created with a libjxl that didn't have this encoder fix yet: https://github.com/libjxl/libjxl/pull/1350 As to what to do with jxl files that have a `jbrd` that expects...

Feel free to make a pull request to get rid of these. I agree that it would be good to get things warning-free (for clang and gcc we make warnings...