Problems in the edge of functions

Problems in the edge of functions

[a2782]

Hi to all!

I\'m using GDB to design a didactic graphic environment over it (as 
DDD). But I have found that, when I put a breakpoint in the first 
machine instruction of a function (push %ebp) and I ask the debugger 
for the information of the frames, GDB \"lies\". An example:

void foo() <-- breakpoint
{
  int b = 2;
}

int main()
{
  int a = 10;
  foo();
}

(in i386 code:
.globl foo
foo:
   pushl %ebp <-- breakpoint
   movl %esp, %ebp
   subl $8, %esp
   movl $2, -4(%ebp)
   leave
   ret

.globl main
main:
   pushl %ebp
   movl %esp, %ebp
   subl $24, %esp
   movl $10, -4(%ebp)
   call foo
   leave
   ret
)

If I type \"info locals\" when the program stops in the breakpoint, GDB 
returns: \"b = 10\" (it should return \"b = 14726457264\"). It\'s evident 
that GDB is looking in the position where it thinks that \'b\' should be 
(but in this position we find \'a\'). The reason why this occurs is that 
the stack has not been updated (the stack updates with the 
instructions \'mov %esp, %ebp\', and \'subl $8, %esp\', but they haven\'t 
been executed). However, if I type \'x /1wx 0x0bfff...\' and \'info 
registers\', I\'ll be able to discover the real state of the stack.

There\'s a similar problem with the instruction \'leave\'.

My question is: is this a bug? Or is there a way of solving it?

Thanks in advance!

Re: Problems in the edge of functions

[Daniel Jacobowitz]

On Thu, Feb 13, 2003 at 05:57:37PM +0000, a2782@dis.ulpgc.es wrote:
> Hi to all!
> 
> I\'m using GDB to design a didactic graphic environment over it (as 
> DDD). But I have found that, when I put a breakpoint in the first 
> machine instruction of a function (push %ebp) and I ask the debugger 
> for the information of the frames, GDB \"lies\". An example:


> If I type \"info locals\" when the program stops in the breakpoint, GDB 
> returns: \"b = 10\" (it should return \"b = 14726457264\"). It\'s evident 
> that GDB is looking in the position where it thinks that \'b\' should be 
> (but in this position we find \'a\'). The reason why this occurs is that 
> the stack has not been updated (the stack updates with the 
> instructions \'mov %esp, %ebp\', and \'subl $8, %esp\', but they haven\'t 
> been executed). However, if I type \'x /1wx 0x0bfff...\' and \'info 
> registers\', I\'ll be able to discover the real state of the stack.
> 
> There\'s a similar problem with the instruction \'leave\'.
> 
> My question is: is this a bug? Or is there a way of solving it?

It's a bug, but there's no real plans to fix it right now; eventually,
better unwind information will probably fix it, and it's not a big
deal.  The locals don't have a value yet, after all.

Sometimes this causes messed up backtraces, too.  It's really less
headache for everyone if you just stop after the prologue :)

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer