The wording here is confusing, because it makes it sound like CMYK/CIELAB will only be applied at the very end of the image transformation pipeline. That would really limit the usefulness of adding these extra color spaces, since doing the manipulation in a different color space than RGB is often the point.
But the linked blog post on GIMP.org[0] words it a little differently:
> What it means for color correctness in particular is that we will now do color conversion only when needed (last-second conversion) and therefore won’t lose information when it could have been avoided. For instance, say you color-pick color from an image: if we were to convert to an intermediate format, before using it on a second image (which may or may not be in another color format), we’d do 2 conversions. Which means more possibility of precision loss. The issue is even more flagrant if the input and output formats are the same (i.e. no conversion should happen at all). And this will be even more a problem when we will have core CMYK backend (we really want to avoid doing a round-trip to an intermediate format with CMYK, which doesn’t have bijective conversion with most other color models, even when working unbounded and ignoring precision issues).
That sounds more like the information of the original color space will be kept as well, and transformations will be applied only when necessary, to avoid lossy roundtrips. Which is not quite the same.
[0] https://www.gimp.org/news/2024/02/21/gimp-2-99-18-released/#...