Automatic backtrace [Was: Re: NetAPP/Samba 3.0/plugs]

[Andrew Bartlett]

David Lee wrote:
> 
> On Tue, 12 Mar 2002, Andrew Bartlett wrote:
> 
> > Danny Braniss wrote:
> > > [...]
> > > [2002/03/12 11:43:14, 0] /vol/src/Samba/samba/source/lib/fault.c:fault_report(3
> > > 6)
> > >   ===============================================================
> > > [2002/03/12 11:43:14, 0] /vol/src/Samba/samba/source/lib/fault.c:fault_report(3
> > > 7)
> > >   INTERNAL ERROR: Signal 11 in pid 1868 (3.0-alpha16)
> > >   Please read the file BUGS.txt in the distribution
> > > [2002/03/12 11:43:14, 0] /vol/src/Samba/samba/source/lib/fault.c:fault_report(3
> > > 9)
> > >   ===============================================================
> > > [2002/03/12 11:43:14, 0] /vol/src/Samba/samba/source/lib/util.c:smb_panic(1124)
> > >   PANIC: internal error
> >
> > You didn't mention that bit...  Can you recompile with --enable-debug
> > and get me a gdb backtrace ('panic action = /bin/sleep 9000' should help
> > in attaching the debugger).
> 
> (Note tangential subject change...)
> 
> This process of obtaining something useful from such internal errors seems
> currently to be in the realm of little-known folklore.  And even when it
> is known, other factors (e.g. time, user-pressure, out-of-hours) can
> conspire against it.
> 
> I wonder whether it might be worth trying to make it (a) easier, (b) more
> automatic, (c) able to help those sporadic, not-instantly-reproducible
> cases?  Something like:
> 
> 1. "configure[.in]" tries to locate "gdb" (or, as a generalised
>    refinement, other possible, known debugging programs);
> 
> 2. "--enable-debug" could be the default for self-build systems (possible
>    issue here if this introduces a significant  performance overhead?);

No performance overhead on modern unix systems - the debug info is left
on the disk when the executable is paged in. 

> 3. If "gdb" (or similar) is found and "--enable-debug" is set, then the
>    default panic action could be to invoke gdb (or other debugger) with
>    the relevant subcommands to perform and log the backtrace.
> 
> Sure, that probably needs refinement.  But for the moment read it for the
> positive possibilities not the worrying nitpicks.
> 
> Promising?  Or an unrealistic non-starter?

Quite promising, and it would *really* help with our debugging.

Andrew Bartlett

-- 
Andrew Bartlett                                 abartlet at pcug.org.au
Manager, Authentication Subsystems, Samba Team  abartlet at samba.org
Student Network Administrator, Hawker College   abartlet at hawkerc.net
http://samba.org     http://build.samba.org     http://hawkerc.net