Dne 15.11.2010 23:04, Matthew Garrett napsal(a):
On Mon, Nov 15, 2010 at 05:01:30PM -0500, Colin Walters wrote:
> On Mon, Nov 15, 2010 at 4:13 PM, Matthew Garrett<mjg59(a)srcf.ucam.org> wrote:
Leaving the retracing at the user's end of things means that the user at
least has a choice in the matter - I'm unlikely to submit any firefox
crashes if I don't have an opportunity to look at what's in the
backtrace. We can make symbols available for local tracing without
forcing the user to download the entirity of the debuginfo, and that
would seem a more reasonable approach.
http://fedoraproject.org/wiki/Features/DebuginfoFS
Agreed, debuginfofs will be better for many scenarios.
Major advantage of the retrace server is that you can get a good
backtraces even from unfresh coredumps. GDB needs access to the binary
and dynamic libraries that were installed at the time of the crash when
generating a backtrace. So if there was an update yesterday that
installed new versions of several packages, it can happen that the
backtrace can no longer be obtained from the coredump locally (without
installing the old packages somewhere else). Retrace server installs
packages that were used at the time of the crash, so the backtrace can
be obtained.
Major disadvantages of RS are that coredump upload takes time, 3rd party
packages can't be easily supported, and users must trust the administrator.