ok, i’m able to reproduce this if i disable opencl. (also with the integration test image)
It’s dependent on the iterations of highlight reconstruction. Since it’s not present with opencl there might be an issue with openmp stuff (there’s some weird behaviour on osx for this)
So you should file an issue for this (https://github.com/darktable-org/darktable/issues/new?assignees=&labels=&template=bug_report.md&title=)
Indeed. If I increase the value “iterations of high-quality reconstruction” slider in the Options tab of Filmic, the red fringes artefact gets progressively worse (minimum artefacts are obtained by setting the slider to zero).
Furthermore, I noticed that by limiting the reconstruction mask to stay just within the borders of the clipped indicators (that is, not crossing those borders), I can actually avoid the appearance of the red fringes. So it seems that the problem manifests especially at the luminosity high-gradient border of clipped areas.