Hi everyone,

A bug was found today in the normalization option of the color calibration module (thanks to neuralizer on pixls.us forum), where the norms of green and blue channels were swaped (it's really just a stupid typo in variables in the code). This bug affects darktable 3.4.0 when using color calibration module, but only with channel mixer normalization enabled, and only the green and blue channels. My bad. Bottom line, you are not affected if you only used color calibration with the chromatic adaptation options, or without channel normalization.

Side note, the bug was found by reading the source code, nobody reported weird effects on images so far.

The fix was pushed today on master and is queued for 3.4.1. However, it means that if you retouched pictures in 3.4.0 with normalization turned on, the visual result might change a bit in 3.4.1. At that point, we (the devs) believe that, on this particular issue, the intent (preserving brightness across the channels change) is more important than the end result (which is currently wrong) and see no point in keeping compatibility over a bug.

*Let me know if that deeply messes-up your current edits, so we can find a solution, and please don't use normalization on green and blue channels until darktable 3.4.1 is released (which should be soon because we already have a load of bugfixes).*

That's also the opportunity to remind you that we need a lot more manpower on testing, shaking and reviewing the darktable master branch. If you or your friends are proficient in C, or just power users who know how to compile from source, please consider shaking the master branch so the releases can be more water-tight. There are just too many bugs we discover after releasing them in "stable" versions only because the master is not tested and shaked enough earlier (especially on Windows). And proof-reading ourselves is very hard, since you only read what you meant, not what you actually wrote, so you don't really see what's written.

Thank you for your understanding, sorry for the inconvenience, and, by the way, happy new year !

Aurélien.


___________________________________________________________________________
darktable developer mailing list
to unsubscribe send a mail to darktable-dev+unsubscr...@lists.darktable.org

Reply via email to