https://bugs.kde.org/show_bug.cgi?id=399923

--- Comment #108 from caulier.gil...@gmail.com ---
For me, the last backtrace is a non sense.

I suspect that the backtrace is corrupted, as GDB is not able to handle the
reality in memory to hack.

Following all the stages in this file where Maik try to fix the crash, i
propose a different approach. My ideas:

- the crash is related to thumbnail creation, and something in thumbnail
storage is broken with face workflow (why, i don't know).
Perhaps the problem is with libpgf codec...

So, to be sure, there is a simple way to switch thumb compression to JPEG for
ex, and to see if something look better. This is just for testing of course.

- the crash is related to OpenCV and OpenCL low level uses. Look this comment
for ex (also relevant of Face management):

https://bugs.kde.org/show_bug.cgi?id=401253#c4

In this case, at least the OpenCV must be compiled without OpenCL support. This
can be certainly done with AppImage, but for distro packages, we need another
way in digiKam implementation to make the code more robust when OpenCL is used
by OpenCV.

Note: this is why i hate all low level performance improvements as OpenCL
framework, which depend in deed of video card device and proprietary modules.
You have no control in application level, and no way to stabilize the user
code. We want a working code first, the improvements if possible without to
degrade all workflow can come later...

And my 10cts €: proprietary video cards modules under Linux are now enough
stable as under Windows. So for me this OpenCL framework is not portable and
safe. And i do not talk about MacOS...

Gilles

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to