Wrong source location in symbols

Wrong source location in symbols

[Fabian Cenedese]

Hi

I seem to have a problem with the code loaction of symbols.

gdb 6.5, gcc 4.1.0, compiled for ppc-eabi, running on WinXP

Here's a snippet from the generated map file:

 .text          0x0000f6f4      0x41c N:/Indel-PPC/Tests/gccext3/bin41/CTaskTemplateClass.603
                0x0000fa28                _TASK_CLASS_NAME_::_TASK_CLASS_NAME_(char*)
...(snip)...
 .text          0x00097bfc       0x14 N:/IMD/Lib41/libnosys.a(write.o)
                0x00097bfc                write

.ctors          0x00097c10       0x98
...(snip)...

.dtors          0x00097ca8       0x10
...(snip)...

.got
                0x00097cb8                PROVIDE (__STARTUP_BEGIN, .)

.startup        0x00097cb8    0x12320
 *(.startup)
 .startup       0x00097cb8       0x6c N:/Indel-PPC/Tests/gccext3/bin41/CTaskTemplateClass.603
                0x00097cb8                _INI_1000__TASK_CLASS_NAME_()

But when I dump the symbol tables I get this:

[1001] T 0xeb20 _ZN17_TASK_CLASS_NAME_6ActionEv section .text  _TASK_CLASS_NAME_::Action()  write.c
[1002] T 0xebdc _ZN17_TASK_CLASS_NAME_D0Ev section .text  _TASK_CLASS_NAME_::~_TASK_CLASS_NAME_()  write.c
...(snip)...
[3549] B 0x98d44 Reg_INI_0101_INOS_TGT section .bss  write.c
[3550] B 0x98d50 Reg_INI_0145_INOS_TGT section .bss  write.c
[3551] B 0x98d5c Reg_INI_0205_INOS_TGT section .bss  write.c
[3552] B 0x98d68 Reg_INI_0160_INOS_TGT section .bss  write.c
...(snip)...
[3567] T 0x99bec _Z27_INI_1000__TASK_CLASS_NAME_v section .startup  _INI_1000__TASK_CLASS_NAME_()  write.c
[3575] T 0x9a530 _Z18_INI_0002_INCO_NETv section .startup  _INI_0002_INCO_NET()  write.c
[3576] T 0x9a5fc _Z18_INI_0150_INCO_NETv section .startup  _INI_0150_INCO_NET()  write.c
[3577] T 0x9a744 _Z27_INI_0002_CSAMDeviceSioLoadv section .startup  _INI_0002_CSAMDeviceSioLoad()  write.c

No matter where they are defined, the symbols get as location the (wrong)
file write.c, so the last valid file from the .text section. I tried with both
-gstabs or -gdwarf-2, but that didn't change anything. I also tried with
forward and back slashes in the paths of the compiled files, no luck.

Where's the code in gdb that would set these locations?

Thanks

bye  Fabi

Re: Wrong source location in symbols

[Daniel Jacobowitz]

On Mon, Dec 11, 2006 at 11:14:52AM +0100, Fabian Cenedese wrote:
> No matter where they are defined, the symbols get as location the (wrong)
> file write.c, so the last valid file from the .text section. I tried with both
> -gstabs or -gdwarf-2, but that didn't change anything. I also tried with
> forward and back slashes in the paths of the compiled files, no luck.
> 
> Where's the code in gdb that would set these locations?

In symtab.c.  If you can build a relatively small testcase, we can
investigate.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Wrong source location in symbols

[Fabian Cenedese]

At 08:41 11.12.2006 -0500, Daniel Jacobowitz wrote:
>On Mon, Dec 11, 2006 at 11:14:52AM +0100, Fabian Cenedese wrote:
>> No matter where they are defined, the symbols get as location the (wrong)
>> file write.c, so the last valid file from the .text section. I tried with both
>> -gstabs or -gdwarf-2, but that didn't change anything. I also tried with
>> forward and back slashes in the paths of the compiled files, no luck.
>> 
>> Where's the code in gdb that would set these locations?
>
>In symtab.c.  If you can build a relatively small testcase, we can
>investigate.

It's not that easy to write a little sample. It may well be the used libs
or the whole tool-chain. That's why I try to find it myself.

I couldn't quite make out in symtab.c a place where the filenames
get set. But I saw in elfread.c, elf_symtab_read that quite some
minsyms got the filename write.c set though they are not in this
source file. I'll see what else I can find.

bye  Fabi

Re: Wrong source location in symbols

[Daniel Jacobowitz]

On Mon, Dec 11, 2006 at 05:48:24PM +0100, Fabian Cenedese wrote:
> I couldn't quite make out in symtab.c a place where the filenames
> get set. But I saw in elfread.c, elf_symtab_read that quite some
> minsyms got the filename write.c set though they are not in this
> source file. I'll see what else I can find.

It's not set per se; it's looked up when needed.  Start with the code
that prints out the dump and work from there.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Wrong source location in symbols

[Fabian Cenedese]

At 11:50 11.12.2006 -0500, Daniel Jacobowitz wrote:
>On Mon, Dec 11, 2006 at 05:48:24PM +0100, Fabian Cenedese wrote:
>> I couldn't quite make out in symtab.c a place where the filenames
>> get set. But I saw in elfread.c, elf_symtab_read that quite some
>> minsyms got the filename write.c set though they are not in this
>> source file. I'll see what else I can find.
>
>It's not set per se; it's looked up when needed.  Start with the code
>that prints out the dump and work from there.

It's exactly the minsymbol filename which is set in above mentioned
function and always enclosed with SOFUN_ADDRESS_MAYBE_MISSING.

I found that many symbols get as filename the most recently used
filename (not only write.c) even if it's wrong. It seems like there
are some file symbols missing/not read or in wrong order or whatever.

Apart from one comparison this filename is only used for displaying,
so I guess I shouldn't rely on its content too much. But still...

I leave that for the moment and write in another post about other problems.

Thanks

bye  Fabi