Crash Reports?

[Akkarin]

Morning all!

I was browsing the core section on the bug tracker and thought "i has idea!". I don't know how feasible it'd be, or even if we'd be able to obtain the desired results.. but would it be possible to have, for example, the map server capture the last few lines of output, and send a trace to us before crashing completely? Under certain strict conditions ofcourse, otherwise we'd end up with reports from script-kiddies attempts at changing their src.

It might give our core devs something concrete to look at instead of half-arsed bug reports with half the info missing. We do it for data collection, could we do it for this too? Just an idea.

[Euphy]

It sounds good in concept, but how feasible is this, exactly? Is there a way to limit reports to unmodified source files, and if so, would that even be useful anymore? (seeing as almost everyone has to modify their source at some point)

[Ind]

its far more evasive than the stat reporter, and at most cases pretty pointless -- there is little to no useful information it outputs prior to crashing.

[Cookie]

Hm, it's not a bad idea but I sort of agree with Ind. In all honesty, the core dumps are really helpful (themselves) especially for printing variables and such for null pointers with gdb. Not only that, this seems slightly more invasive.

[Brian]

Any crash info it sends prior to actually crashing would not be as valuable as the dump after crashing, right?

What if we add a script (Bash, Perl, PHP, something) in /tools that users can run?

- it would search ../ for core dumps

- run them through GDB to get 'bt full'

- send that to us (via email, HTML Post, or the easiest method we can think of)

- would it be necessary to send copies or blocks of certain source files so we can see what line numbers are being referenced?

- optionally delete the core dump after (so it doesn't resend it next time)

or we can do the filtering on our end to ignore duplicates

The script would be completely optional and would require them to run it in order for it to send anything.

[xazax]

What I used to use with my server was a script which detected if the server crashed and issues the following command:

gdb -q -s $1 -c core -ex 'bt full' -ex 'quit' >> crashdump.log

where $1 was the name of the server.

Of course the script executed ulimit -c unlimited before it started the servers.

I think it is pretty easy to implement some shell script for automatic dumps, the problem might be to make it platform independent, I don't know how big the differencies are across linux distributions for instance.