abrt wishlist

[Tom Tromey]

>>>>> "Jan" == Jan Kratochvil <jan.kratochvil at redhat.com> writes:

Jan> GDB on client side would need something like readonly NFS-like
Jan> service to load the .debug files byte-wise.  And this NFS-like
Jan> service network protocol must be signed by Fedora project like the
Jan> current rpms are.

Jan> Then the fast operation of GDB on the client depends on these known
Jan> issues:

Jan>  * http://fedoraproject.org/wiki/Features/GdbIndex which is now
Jan>  present in F14 updated packages (but it still was not present in
Jan>  F14 GA packages).

Jan>    * Recheck THe .debug files reading via .gdb_index is really
Jan>    optimal for the network protocol.

I assume it isn't.  The core of the index is a memory-mapped hash table.
GDB will access it in essentially random order.

Jan>  * Fixing GDB to not ever touch .debug files for libraries not going to be
Jan>    used in the specific backtrace at all.

This seems like a good idea.  Maybe GDB could detect that the debug info
is remote and defer the reading only in this case.

Even if we don't read the debug info, we probably still want to read any
python code associated with the library.

Another idea we could implement is do index queries over the net, say
using a remote SQL server.  This means more round trips, but less data
to download.  Maybe this isn't really the right tradeoff -- but the
point is that we have a lot of leeway to change GDB, we don't have to
try to work around it.

Tom