Re: [avr-libc-dev] Re: [avr-gcc-list] dwarf2 information, is this a bug?

[Björn Haase]

Eric Weddington wrote on Montag, 17. Oktober 2005 17:44 :
> Björn Haase wrote:
> > In reply to a message from July this year posted on avr-libc-list:
> > Yes I think, Thorsten, that you found a bug.
> >
> > It seems that the binutils mainline does no longer assemble correctly the
> > dwarf2 informations. E.g. I observe that avr-addr2line no longer finds
> > the correct line number for newly assembled files.
> >
> > When using the rev. 2.16 avr-addr2line on files assembled with 2.14 the
> > correct line numbers come out. So addr-2line seems to be functional in
> > 2.16.
> >
> > My present estimation is that the problem is probably with gas or bfd.
> >
> > Bjoern.
>
> Hi Björn,
>
> Do you think the problem first showed up in 2.16? Would it be better off
> to stay with 2.15 until this issue is fixed in HEAD?

No. Last week I indeed had observed the same issue Torsten has reported. Today 
I have analyzed the issue and found out that the problems with 2.16 showed up 
for my applications only in conjunction with my new linker optimizations. It 
seems that Torsten's issue has been fixed meanwhile. I found out where the 
problem with my linker relaxation patch were and I have fixed the issue this 
afternoon. I will soon post a follow-up patch of the linker-relax 
optimization.

My suggestion would be to be even less conservative: I'd even suggest that 
within two weeks time you could ship *mainline* binutils that includes linker 
relaxation: I observe that the code size decrease might justify it and in 
case that there are unexpected new problems, you simply would not need to 
switch --relax on.

BTW: You are now employed by Atmel, Eric?

Yours,

Bjoern