Drastic change in colors for minute change in color matrix 1

[10110111]

I was trying to generate valid DNG files from my phone camera to be able to process in RawTherapee (checking on Ubuntu 18.04 RawTherapee 5.3). For some reason, it appears that if I go from the value of TIFF field with TIFFTAG_COLORMATRIX1 of

{1,    0,   0,
 0,  2.022, 0,
 0,    0,   1}

to

{1,    0,   0,
 0,  2.023, 0,
 0,    0,   1}

the colors in RawTherapee change drastically: from greenish tint to purple. See the following screenshots:

I do admit that the files are not yet fully correct: e.g. AsShotNeutral field is currently a stub (1,1,1), but such a sensitivity of RawTherapee to minute changes in a diagonal element of the color matrix sounds suspicious, so I think it's a problem somewhere in RawTherapee.

The relevant test DNG files can be downloaded here (14MiB XZ archive).

[Beep6581]

Maybe related to #5496

[Entropy512]

Hmm... Yeah, I think #5496 may have a couple of subcomponents (most notably, your question of why ColorMatrix2 is being used when Auto-Matched Camera Profile is active). Some of those might be contributing?

[iliasg]

Interesting is what happens after passing the two DNGs through Adobe DNG Converter !! The colormatrix1 gets normalized to 0.4946 0 0 0 1 0 0 0 0.4946 - 0.4943 0 0 0 1 0 0 0 0.4943 and the rendering looks more reasonable (and almost identical for both DNG_1s) ;)

Also interesting that the tint value is almost same as the central multiplier in colormatrix1 !!

Anyway .. looks like both the greenish an purplish renderings of the prototype DNGs are wrong because RT's WB calculations "crash" on the tint limits (10.0 or 0.02)

[Beep6581]

To be tested whether this can still be reproduced in @Desmis 's white balance branches.