Re: getting the line of a function when walking the stack with libbfd

[Vincent Torri <vincent dot torri at gmail dot com>]

On Tue, Mar 20, 2012 at 7:39 AM, Alan Modra <amodra@gmail.com> wrote:
> On Tue, Mar 20, 2012 at 05:39:08AM +0100, Vincent Torri wrote:
>> On Tue, Mar 20, 2012 at 5:33 AM, Alan Modra <amodra@gmail.com> wrote:
>> > On Mon, Mar 19, 2012 at 07:03:04PM +0100, Vincent Torri wrote:
>> >> But all of them do not find the correct line where the function is
>> >> called. It seems that the line is actually the one finishing the
>> >> function (that is a 'return' call or the closing curly bracket)
>> >
>> > Does addr2line work?
>>
>> I actually don't know how to use the addr2line program (I have to pass
>> an address and I don't know which value i have to use)
>
> Disassemble some of your binary with objdump to find call sites, plug
> in those addresses to addr2line. ?Experiment a little with other
> addresses near the calls.

ok, i'll try that. There's something I didn't mentioned (but you might
have guessed it) : the tool I'm writing is for Windows programs
compiled with MinGW. I already have something working with vc++.

Also, the code [2] and [3] are working like my tool (it's almost the
same code, anyway).

> There are many possibilites why your stack trace isn't giving you the
> lines you expect. ?eg.
> - You may not realise how stack traces work, ie. that they give
> ?return addresses, not call site addresses, or

that's quite possible :-) I've tried to write that tool by searching
examples on internet. But it's certainly possible to do what I want
as, for example, gdb is doing it (but is not using libbfd for stack
trace, iirc).

> - You may not have any line info in the binary, or

how can I check that ? Note that I compiled with -g, so I guess that I
have some debug info in my binary :)

> - The line info may be faulty, or

how can I check that ?

> - your binutils BFD may be buggy.

I've compiled the latest release of binutils and statically linked
libbfd to my tool.

Of course, I can provide the code (it's small, around 550 lines of
code), if you want to look at it :)

regards

Vincent Torri