[nicknezis]

With a PQ or HLG display profile, lighttable and filmstrip thumbnails are much lighter than the darkroom view, as seen while testing #22669. The open image's own filmstrip thumbnail looks right.

Cause. Since 4b0878d the pipe encodes PQ with 1.0 at reference white (203 nits) and applies the HLG OOTF itself, through dt_hdr_encode(). The thumbnail display transforms (transform_srgb_to_display, transform_adobe_rgb_to_display) still go straight through LittleCMS to the built-in PQ/HLG ICC profiles, whose shapers (_PQ_fct(), _HLG_fct()) put PQ 1.0 at 10000 nits and have no OOTF. The image open in the darkroom escapes this because its filmstrip thumbnail is drawn from the preview pipe's backbuf (dtgtk/thumbnail.c), which colorout has already encoded.

Fix. For PQ and HLG display profiles, the sRGB/AdobeRGB display transforms now end in linear light (via dt_colorspaces_linearize_profile(), as colorout does), and a new dt_colorspaces_transform_to_display() applies dt_hdr_encode() before writing 8-bit BGRA. Lighttable/filmstrip (dt_view_image_get_surface()) and the tethering live view use it. Other display profiles keep the plain 8-bit transform, so SDR output is byte-identical. The display2 transforms get the same treatment for consistency, though nothing uses them at the moment. The ICC profiles themselves are unchanged, so exports and imports are not affected.

Checked. Builds without warnings (Ubuntu 24.04, GCC, OpenCL off); the hdr unit tests pass. A scratch program linked against libdarktable, run headless with --conf ui_last/color/display_type=..., pushed an AdobeRGB gray ramp through the old and new thumbnail transforms and compared with dt_hdr_encode() of the same linear values (display code, green channel):

AdobeRGB code	PQ old	PQ new	PQ pipe	HLG old	HLG new	HLG pipe
0	0.055	0.000	0.000	0.000	0.000	0.000
117 (18% gray)	0.816	0.412	0.411	0.659	0.435	0.436
186 (50%)	0.925	0.510	0.509	0.867	0.631	0.632
255 (white)	1.000	0.580	0.581	1.000	0.749	0.750

HLG P3 gives the same grays; saturated colors differ only by primaries. With the sRGB display profile, old and new output are identical.

Not checked. I haven't looked at the thumbnails in the GUI on a PQ or HLG screen, nor the tethering live view.

Release notes. None added: the mismatch came in with 4b0878d in this cycle, so no release had it. Happy to add one if you'd rather.

This change was written with AI assistance (Claude Code) and reviewed by me.

[TurboGit]

Works for me in all views, lighttable, darkroom, slideshow and print. Good to go, thanks!