Code review comment for lp:~vanvugt/mir/fatal-error

This negates the benefits of previous work to ensure proper shutdown in case of server problems, since many (most?) of our exceptions, especially in the platforms, are not recoverable and hence could be changed to mir::abort(). From that perspective this MP causes a regression for https://bugs.launchpad.net/mir/+bug/1189770 .

Like last time we had this discussion, it's a matter of what we deem to be more important: user experience or crash information. Ideally we want both, and there are some avenues worth exploring that may allow us to achieve that middle ground. For example we could get the core with 'gcore' and pipe that to whatever command/file is in /proc/sys/kernel/core_pattern, mimicking a crash, and then graciously shutdown mir.

On the other hand, if the init system is able to bring USC/unity8 back up properly after an abort (which is probably the case) then this is a problem mostly for developers (e.g. crashing a test server and not being able to change VTs) not for normal users. So, perhaps the best way forward is to be able to configure what mir::abort() does. For example, it could std::abort by default, or optionally throw + gcore to a file.

"Needs discussion"