Re: New stack trace code

[Bryce McKinlay <mckinlay at redhat dot com>]

Andrew Haley wrote:

> Bryce McKinlay writes:
>
> > Try -Dgnu.gcj.runtime.NameFinder.use_addr2line=false
> > 
> > addr2line is certainly too slow when an application calls it frequently, 
> > but I think maybe we should wait until we do some performance tests on 
> > the dwarf2 reader
>
> That depends on when we might get the dwarf2 reader.
>
> > before deciding whether it is neccessary for an option that
> > disables line numbers globally.
>
> Why?  We can always take it out again.  If it just so happens that the
> dwarf2 reader makes it unnecessary, then we will rejoice.
>
> At the present time logging is painful because of the calls to
> addr2line, and the ability to disable it is an enhancemant.

Yes, but we already have a system property to do this! Do we really need 
to add another one?

> > Also, the overhead of getting line numbers for interpreted frames
> > is small, so I don't think it makes sense to disable for that case
> > as well.
>
> Sure, but if an application has no need for line numbers, then it
> makes no sense to get them, no matter how small the overhead.
>  

I think line numbers are useful for any application. Unless your 
application has no bugs, eventually someone will want a useful 
stacktrace. This is especially true for an application server, where 
faults may be intermittent and not easily reproducable in a debug 
environment! As long as the overhead isn't too significant, line numbers 
should always be enabled when possible, just as they are in other VMs.

Bryce