Re: Partial linking problems on ELF32 PPC

[Ranjit Deshpande <RanjitDeshpande at AirgoNetworks dot com>]

Daniel,

Thanks for your pointer. I've verified that your gdb patch works both with
partially linked objects and with shared libraries. Is this patch going into
the gdb tree ?

Since you mentioned that you did not complete your patch to objdump and
readelf, could you please post it ?

Thanks,
-Ranjit

On 12/28/02 8:14 AM, "Daniel Jacobowitz" <drow@mvista.com> wrote:

> On Sat, Dec 28, 2002 at 01:42:58AM -0800, Ranjit Deshpande wrote:
>> Hello,
>> 
>> I've run into this problem while using partial linking with multiple object
>> files where the DWARF2 .debug_info section incorrectly tells me that a
>> string is at offset X in the merged .debug_str section. I do think that the
>> correct offset is present in rela.debug_info but none of the tools (objdump,
>> readelf or gdb) use this information.
>> 
>> Included below is the output from readelf -wi ./foo.o. I've chopped off the
>> uninteresting bits. Notice the DW_AT_name attribute for DW_TAG_compile_unit.
>> It shows the correct file name for the first CU, but the latter CU's are
>> screwed up. I have 3 objects: main.o main1.o main2.o and the link command
>> is:
>> 
>> $ ld -r -o foo.o main.o main1.o main2.o
>> 
>> What I have noticed is that the .debug_info section contains offsets into
>> .debug_str that are no longer valid after the merge.
>> 
>> I claim complete ignorance of how all of this works and the fact that the
>> rela.debug_info shows the offset into .debug_str correctly makes me think
>> that the offset is not completely lost after merging the .debug_info and
>> debug_str sections from the 3 objects.
>> 
>> However readelf and objdump both do not take the rela.debug_info into
>> account.
> 
> Now this is peculiar.  I swear that I fixed this - it wasn't in 2.13.1,
> but it should ave been in the current snapshot.  No, looking back at it
> it looks like I only fixed the abbrev offsets.  Apparently I forgot to
> finish what I started.
> 
> The correct fix for this is to call
> bfd_simple_get_relocated_section_contents from readelf (and objdump).
> 
>> And being lazy, I wonder if Daniel Jacobowitz is lurking on this list so
>> that I might ask him if he is aware of this and has dealt with this in the
>> gdb code. From my extremely limited understanding of the gdb code, I don't
>> think that he has but I could be wrong. GDB does not find source files.
> 
> I posted the patch to fix this to the GDB list some time ago and am
> still waiting for comments on it.  Check the archives of gdb-patches,
> Oct. 11th, "[RFA] Relocate debug information in object files".
>