Reading alex post about Debugging without symbols I remebered some debugging magic I did some time ago: I had to debug a core file generated by a dynamically loaded module on a Solaris without debugging info, having only a non stripped module on another machine. Lot of magic with nm and add-symbol-file. Thinking about what we can do for GNOME…
- Users running jhbuild or garnome are ok, because they have debug symbols and provide useful backtraces
- Users with a vendor instalation are not ok, because they don’t have debug symbols, and asking them to download a bunch of MiBs is not acceptable
So this is my propossed solution for these backtraces without debug info:
- Add to bug-buddy a new field with the md5sum of the crashing binary (and libs?)
- Set up a big machine with jailed enviroments (one per distribution or one per distro-gnome-bins-and-libs pair)
- Sent the backtrace to that machine
- Run on it gdb against the matching env (chrooting to it)
…and we will have a beatiful stact trace. Some more open questions:
- Should bug-buddy directly sent the bactrace to the “debug server”,get the answer and send final report to bugs@gnome.org? Or should the magic procmail at bugs@gnome.org get this info from the server
- Can we get this “big” machine? Maybe Intel? sri?
- Do we have volunterrs for se5tting the machine up (installing all jailed distros) and maintaing (installing new distros)