Project

General

Profile

Bug #9520

crash in lowlight module during export due to eq producing NaNs

Added by Igor Kuzmin almost 6 years ago. Updated about 5 years ago.

Status:
Confirmed
Priority:
Medium
Assignee:
-
Category:
-
Target version:
-
Start date:
07/18/2013
Due date:
% Done:

10%

Affected Version:
git development version
System:
all
bitness:
64-bit
hardware architecture:
amd64/x86

Description

I've come across one specific combination of image+history stack that produces segfault in lowlight.c during HQ export on darktable 1.2.2 (tested both release and current HEAD of darktable-1.2.x branch).
raw file: http://paraf.in/junk/IMG_5957.CR2
minimized history stack that still produce a crash: http://paraf.in/junk/IMG_5957.CR2.xmp

darktable_bt_258E0W.txt Magnifier - backtrace (48.2 KB) Igor Kuzmin, 07/18/2013 09:06 AM

darktable_bt_ZF96ZW.txt Magnifier - backtrace with -O0 and no OpenMP (41.4 KB) Igor Kuzmin, 07/18/2013 01:12 PM

History

#1 Updated by Igor Kuzmin almost 6 years ago

  • % Done changed from 0 to 10
  • System changed from other GNU/Linux to all
  • Status changed from New to Confirmed

Just tested on my Mac with 1.2.2 release DMG, I get the same crash, so it's not specific to my Linux system. hanatos tried it yesterday on master - it didn't crash, so the bug is most likely only present in 1.2.x branch and was fixed in master. Some cherry-picking would be nice...

#2 Updated by Igor Kuzmin almost 6 years ago

hmm, what's strange though is that I just checked and there's no code difference in lowlight.c between master and darktable-1.2.x branches. maybe hanatos just did non-fullres non-HQ export (it doesn't crash in that case)...

#3 Updated by Igor Kuzmin almost 6 years ago

I've recompiled with -O0 and without OpenMP. Crash is still there, I've attached the backtrace.

#4 Updated by Igor Kuzmin almost 6 years ago

  • Subject changed from crash in lowlight module during export to crash in lowlight module during export due to eq producing NaNs

just to keep record: equalizer module (atrous) was identified as an original source of the problem - it produces NaNs

#5 Updated by Tobias Ellinghaus over 5 years ago

  • Status changed from Confirmed to Incomplete
  • % Done changed from 10 to 20

Can you still confirm this with git master?

#6 Updated by Igor Kuzmin over 5 years ago

I remember asking hanatos about this bug - he said he did come up with a fix, but it wasn't good enough for him to be satisfied so the patch wasn't committed to git. I don't remember if it applies to git master.

#7 Updated by Roman Lebedev about 5 years ago

  • bitness set to 64-bit
  • Status changed from Incomplete to Confirmed
  • % Done changed from 20 to 10
  • Affected Version changed from 1.2.2 to git development version

Still crashes with current git master

process._omp_fn.0 () at /home/lebedevri/darktable/src/iop/lowlight.c:120
120 return lut[bin1]*f + lut[bin0]*(1.-f);

Also available in: Atom PDF