[RafatH0ssain]

If a raw reader reports a white point at or below the black level, commit_params() in rawprepare computes div = white - black <= 0. The image then renders black or inverted, and nothing tells you why. This happened with LibRaw on the Canon EOS R50 V (white 90 or 148, black 2047; upstream fix in LibRaw/LibRaw#875).

As suggested in #22597, this adds a cheap check in rawprepare's commit_params():

• if raw_white_point <= max(raw_black_level_separate[]), the module uses UINT16_MAX for the scaling and for the white point it passes downstream, so the image is dark but not inverted
• the module shows an "invalid white point" trouble message (also printed to stderr), so the user can set a correct value in the module
• the message only appears while the module is actually active, and it is cleared once the white point is valid again; the clear happens before _image_set_rawcrops() so it can't wipe the existing crop-parameter message

This doesn't touch camera support. The R50 V modelMap entry from the issue is left out on purpose, per the discussion there.

Tested on macOS (Apple Silicon, Homebrew, no OpenMP):

• builds with -Werror (rawprepare.c clean); ctest 40/40 pass
• darktable-cli on mire1.cr2 with the rawprepare white point set to 100 (black 1030): before, the image is essentially black (mean 0.0003) with no message. After, it renders darker but not inverted (mean 0.18 against 0.41 normally) and logs Trouble: [raw black/white point] 'white point not above black level, using maximum'
• darkroom (GUI): opening the same image shows the "invalid white point" banner on the raw black/white point module. Resetting the white point to the camera value (13600) clears it and the image renders normally, and undo brings the banner back
• integration suite: 169/191 pass. The other 22 fail the same way on this machine without the patch, with identical max dE per test (isolated pixels, avg dE ~0.001), so they look like platform differences. None of the 191 tests hit the new fallback

AI disclosure: the change was drafted with AI assistance (Claude Code + an OpenCode model) and reviewed by me.

Related: #22597

[jenshannoschwalm]

Please have a look at the dev-doc about thread sharing gui data :-) You also will find examples in other modules ...

[jenshannoschwalm]

Squash both commits and then it's good.

[TurboGit]

Thanks!