please ship & setup a core_pattern handler

[Harald Sitter]

Hola.

I've changed kcrash to default to raise() signals to core_pattern
handlers instead of invoking drkonqi directly, by default. This should
help us produce more reliable backtraces by reducing the racing
potential between different threads in the crashing application.

To help implement this it'd be grand if all Linux distros could make
sure to have a core_pattern handler set up out of the box. systemd's
coredumpd is natively supported by drkonqi, if you want to use
something else you'll have to implement your own UI layer on top.
Using ABRT or apport is also totally fine. So long as whatever you are
using ties into the kernel's core handling support you are good.

This replaces the previous opt-in behavior of KCRASH_DUMP_ONLY=1.

For the time being direct drkonqi invocation is still an option but I
think longer term we'll have to find ways to move away from it, it's
just too unreliable. In the past month we've lost 1/10 of the
megarelease alpha crashes to process restart races and an unknown (but
noticeable) amount to fault races. I'm hoping by leaning on the
kernel's core dumping support more we can get these numbers way down.

https://man7.org/linux/man-pages/man5/core.5.html
https://kde.haraldsitter.eu/posts/drkonqi-%EF%B8%8F-coredumpd/

HS