Auto-generate backtraces & crash reports

[vorner]

It would be good if whenever a process crashes, a crash report with some useful information is produced and saved somewhere, noted in a log and the admin is asked to send it. The idea is to use whatever is available in the system to get the information, for example use gdb to attach to the process and get the information or use some built-in methods. The crashing process could have a handler for registered all reasonable fatal signals and call a shell script with it's PID and name and the script would write down system information, call gdb on the PID, produce the backtrace, etc…

It might make sense to register the signals for python too (in case it crashes in the wrapper code) and make sure any uncaught exceptions are caught at the highest level and written down to a file.

I attach a file which does that (from one of my own projects), it is slightly different, it produces the backtrace to the screen and leaves the gdb running, but it can illustrate the idea. While this relies on having gdb installed on the system and having debug information compiled in (which we could enable by default), it would at least increase the chance of getting useful bug report and make it easier for a user to generate them.

[shane]

Milestone Y4 Usability deleted

[tomek]

Milestone renamed