Rawtherapee: Core dump RL convolution followed by Contrast+

[ClaesGitHub]

In RT's file browser:

1. Select image (either RAF or TIFF)
2. Apply Neutral
3. Open

Now in RT's Editor:

1. Zoom to 100%
2. Detail tab

1. Activate Sharpening -> RL Deconvolution

2. Set Damping to 0

3. Activate Contrast by Detail levels

4. Contrast+

This results in SIGSEGV. See debug dump below.

AboutThisBuild.txt
log.txt

Just tell me if you need additional information.

Sincerely,
Claes in Lund, Sweden

[Beep6581]

Reproduced. I had to click Contrast+ several times, then it crashed.
Using http://rawtherapee.com/shared/test_images/colortarget/FUJIFILM%20X-T1%20daylight.raf

[Beep6581]

Reproduced also using amsterdam.pef, same procedure.

[heckflosse]

Reproducible in dev, but not in pixelshift_fixes, which is 26 commits behind dev.
This at least reduces the number of bisect steps ;-)

[heckflosse]

Bisected to https://github.com/Beep6581/RawTherapee/commit/b42a45b4817ca838d0d8490638da9f94f35d9b71

ping @agriggio

[agriggio]

I'll look into this once I'm done with #4479
But please feel free to work on it :-)

[ClaesGitHub]

ping @agriggio

CIao Alberto,

I have a hunch that it is the
_Set Damping to 0_
that is the culprit here, i.e. if
Damping > 0, there is no problem (?).

Cordiali Saluti,
Claes a Lund, Svezia

[heckflosse]

@ClaesGitHub Setting the damping to != 0 avoids the crash because it implicitely avoids negative Lab L values before the next processing step. Bisecting clearly shows the responsible commit and RL deconvolution code was not changed with this commit afaik.

[agriggio]

yes sorry, it's a side effect of unbounded processing. I knew it would happen sooner rather than later :-)

[agriggio]

@ClaesGitHub can you try again with the latest dev? I pushed some changes that might have made this go away -- I just tried and was unable to reproduce

[ClaesGitHub]

@agriggio
Ciao!
Thank you. I have just checked the latest dev and I can _not_ reproduce the error.
/Claes

[heckflosse]

@agriggio Fix confirmed :+1: