On 28 January 2013 22:04, Jonathan Wakely wrote:
> On 28 January 2013 22:00, Miles Bader wrote:
>> I've noticed that if I compile my C++ programs using (a recent Debian
>> snapshot of) the gcc trunk [0], and try to debug the resulting
>> program, then I've noticed (there are no doubt other issues):
>>
>> (1) gdb doesn't seem to know about most local variables / function
>> parameters: trying to print them by name doesn't work ("No
>> symbol ... in current context); "i loc" and "i args" show nothing.
>>
>> (2) gdb can't find many functions in my program: referring to them
>> by name doesn't work,and searching using "i fun ..." with an
>> appropriate pattern gives nothing.
>>
>> (3) When printing backtraces, etc, gdb spits out warnings like:
>> warning: (Internal error: pc 0x429e7a in read in psymtab, but not in symtab.)
>>
>> (4) Backtraces sometimes seem to be missing frames that I don't
>> think are due to inlining/tail-calls/whatever.
>>
>> I.e., things are fairly clearly "messed up."
>>
>> This behavior seems to be the case no matter what optimization flags I
>> use (from no -O flag at all, to -O3; I always use -g).
>>
>> If I recompile my program using gcc 4.7, then gdb works as expected
>> (local variables, functions, no warnings spat, etc). This is true
>> regardless of optimization level.
>>
>> Is there some change in trunk gcc that requires a newer gdb or something...?
>
> I was seeing those same issues, e.g.
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55041#c6
>
> But the latest comments on that bug indicate it's no longer a problem
> with current trunk (I haven't verified that myself yet, and you're
> using a pretty recent snapshot)
Oops, I mis-remembered the PR trail, the thing that's fixed is
something else. The comments on the PR (and a change to the 4.8
release notes) say you need a new GDB or to disable DWARF-4 output.
