(on a side note, .cheat .crash doesn't produce a stacktrace here)
LuaUI (0.74b3) may cause crashes when hosting
Moderator: Moderators
Nothing in infolog.txt?
It doesn't necessarily have to be written in php, but that seemed easy because then the 100M debug executable can just sit at a server (though you could make a perl cgi script or something like that too with the same effect).
More details:
I now convert stacktraces by the following command:
Break it in pieces to see what it does, it needs the debug exe and an infolog.txt with a stacktrace in it.
The debug executable for 0.74b3 (the spring.exe referenced in the above command) is available at: http://www.osrts.info/~tvo/spring/rc/sp ... -debug.exe (this is a self extracting 7z archive, the actual debug spring.exe in it is over 100M (max debugging enabled)). It is exactly the released executable apart from one call:
It doesn't necessarily have to be written in php, but that seemed easy because then the 100M debug executable can just sit at a server (though you could make a perl cgi script or something like that too with the same effect).
More details:
I now convert stacktraces by the following command:
Code: Select all
egrep '\([0-9]+\) [A-Za-z:\/. ]+ \[0x[0-9A-F]{8,8}\]' infolog.txt | \
awk -F '\[|\]' '{print $2}' | xargs addr2line -f -C -e spring.exe | \
sed 's/^Z/\n_Z/g' | c++filtThe debug executable for 0.74b3 (the spring.exe referenced in the above command) is available at: http://www.osrts.info/~tvo/spring/rc/sp ... -debug.exe (this is a self extracting 7z archive, the actual debug spring.exe in it is over 100M (max debugging enabled)). It is exactly the released executable apart from one call:
Code: Select all
strip spring.exeYeah, infolog is empty (ends at the ".crash" command). I haven't looked at it myself, but is the stackwalking code disabled on Linux?
edit: never mind, judging from the absence of a crash-handler in System/Platform/Linux I think I know the answer to that one
edit: never mind, judging from the absence of a crash-handler in System/Platform/Linux I think I know the answer to that one
Last edited by Kloot on 09 Jul 2007, 01:32, edited 2 times in total.
You can make it a lot faster by buffering the hexadecimal addresses and only calling addr2line once, that's what I did in C++ for the sync debugger. (because, as you may have noticed, it's the loading time of addr2line that makes it slow, it needs to load 100M debug symbols after all...)
Indeed, I didn't write a linux stacktrace thing. For source distrib it doesn't make sense after all, could just use it for the debs etc., but then there needs to be a foolproof way of identifying the binary (ie. putting md5sum of the binary in the infolog or something).
Indeed, I didn't write a linux stacktrace thing. For source distrib it doesn't make sense after all, could just use it for the debs etc., but then there needs to be a foolproof way of identifying the binary (ie. putting md5sum of the binary in the infolog or something).
Updated the patch link:
http://trepan.bzflag.bz/spring/luaui_co ... .74b3.diff
http://trepan.bzflag.bz/spring/luaui_co ... .74b3.diff