Being nice to testers - mandating debug packages

[Phillip Susi]

It isn't that I don't think having the debug symbols around isn't a good 
idea -- it is; I just don't see how testers will use them, except with 
KDE apps that pop up the nice bug window with the backtrace.  Gnome 
doesn't do that does it?

In any case I am not sure just how useful the gui backtrace is, and it 
certainly doesn't apply to command line apps.  It seems to me that it is 
more helpful to send the core dump to the developer than just the 
backtrace, then let him load up the debug symbols with the core dump.

I just converted a package last night to build a -dbg version so I could 
debug it.  It didn't take much effort to do and the only down side to 
having the -dbg package is it uses some more space on the mirrors.

David Nielsen wrote:
> fre, 13 01 2006 kl. 11:20 -0500, skrev Phillip Susi:
>> I've been thinking about this lately myself, but I don't see this being 
>> of any use at all to testers.  Testers aren't going to get the debug 
>> symbols and fire up gdb for a debugging session, especially on the 
>> optimized binaries.  The only place I can see this being used is by a 
>> developer who asks the tester to send him the core dump, then he will 
>> need the correct symbols to analyze the core dump.
> 
> Testers will most certainly get the debugging packages, as long as they
> know they are there - which is why I maintain a small bugfiling how-to
> on the Forums btw. At least a significant amount of testers will, the
> rest will be just as loud as they are now.
> 
> We already have good guides to obtaining traces on the various wikis
> around important projects and it seems to be policy for developers to
> link to it in bugreports if they get set to NEEDINFO.
> 
> I will readily admit that my ability to read backtraces is somewhat
> limited, but how big an impact on usefulness is the optimized split
> debug, they seem to be good enough for the RedHat developers at least?
> Besides some bugs happen only optimized packages since GCC nor
> application developers are perfect and bad behavior is to be expected
> from time to time.
> 
> Even assuming your argument that testers will simply not get the debug
> packages, does it then not for the sake of creating the best possible
> product make sense to enforce debug build by default during at least
> part of the development cycle. Crashing applications require proper
> fixing and as testers of the platform we should expect it to take some
> effort sometimes, at least if we want the bug to go away. Yes this will
> bloat the install and cause some slowdowns but if this is clearly
> explained to the users like e.g. FreeBSD does for development cycles of
> their base system and kernel, I don't see a major problem, the slowdown
> will go away in final release as will the extra symbols.
> 
> - David Nielsen