And finally, the bug that took me LITERALLY a day to properly reproduce it in a CI, even if I’m yet able to identify why. This bug emerged when I tried to compile gobject-introspection.
When using a combination of ThinLTO and -march=x86-64-v3 -fstack-protector-strong. I get:

samu: job terminated due to signal 11: /home/build/build/tools/g-ir-scanner --output=gir/GLib-2.0.gir [...]

My hypothesis is that the linker segfaults when handling -lpython3.6m.
This issue emerged between d7e201c and 5640962 too.
Here are the logs: before it worked, but not after.
The only difference between the two builds is the commit hash of python-cmake-buildsystem.

I’m done here for this week. Took me too much of my time :/

Originally posted by @concatime in #293 (comment)