In recent master builds, I often have instant crashes of Darktable when switching from inpaint-opposed to segmentation.
Also, when I start with a fresh clean history on an image, segmentation appears to work. But if I start dragging sliders like the ‘candidating’ sliders, Darktable often crashes.
I have no debug output anymore when starting Darktable with -d all, the window just stays black. Something in my msys builds I guess (or the change to ucrt?).
So I can’t give proper debug info. But before making an issue on Github, I thought about asking ‘am I the only one?’.
I’m in for it. Unfortunately noone reported if they could reproduce. There also seem to be some PRs pending concerning segmentation based I was thinking about to wait for.
Maybe doing a forced clean build (git clean -d -f -x) could help?
Some crashes have been diagnosed and fixed; there’s also a new debug flag, -d roi, to help with a certain class of problems, which was used (I think) to find the problematic code.
I guess this is coming a bit late but I’ve just installed the latest windows weekly build 4.1.0+681~g19825f35c
and it’s crashing as descibed in the OP, also with no warnings - it just quits.
Just mentioning in case it helps!
As described in the last post, there is a fix for the issue via the PR. The PR is not fully ready to merge because of also trying to fix a mask issue too.
@hannoschwalm I just compiled master as of today, it seems the segmentation crashes are back (or another issue appeared).
Moving the candidating slider makes Darktable go poof.
edit: I’ve also noticed some crashes while holding ALT and clicking on a duplicate to preview it (no, that duplicate does not have segmentation-recovery enabled ). I don’t know if it’s related or not. After it crashes, I can often restart Darktable and try again, and it works. But segmentation-based messing with the candidating slider is a sure way to kill Darktable now.