Re: [PATCH] better assembly level debugging

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Francois" == Francois  <rigault.francois@gmail.com> writes:

Francois> (continuing http://sourceware.org/ml/gdb/2011-05/msg00009.html)
Francois> I've made a proof of concept patch to pretty print addresses. I
Francois> believe it can already greatly improve the task of reading assembly
Francois> code, although it is not yet ready for delivery, I'd love some
Francois> feedback.

Francois> I'm finally able to display:
Francois> => 0x400680 <main+4>:   sub    $0x10,%rsp
Francois>    0x400684 <main+8>:   mov    0x1003fd(%rip),%rax        # 0x500a88 [<wprintf>]
Francois>    0x40068b <main+15>:  mov    %rax,-0x10(%rbp)
Francois>    0x40068f <main+19>:  lea    0x106(%rip),%rax        # 0x40079c [L"foo 42"]

Francois> Examining the stack:
Francois> 0x7fffffffdd00 [<wprintf>]:     0x7ffff7371920 <wprintf>
Francois> 0x40079c [L"foo 42"]

Looks nice.

Francois> - a gdb.probe_address Python function is called by the print_insn
Francois> routine to check if operands can be pretty printed. If yes, the # is
Francois> added on the right of the instruction and the print_address_func is
Francois> called
Francois> - a gdb.print_address Python function is called by
Francois> print_address_symbolic if no symbolic address could be created for
Francois> this address by the C code.

I am not a big fan of this approach.  It means that different Python
modules that want to install printers will conflict.  It should probably
be handled some other way instead.

Francois> - the code is not safe yet and I did the changes only for i386-dis.

Changes to opcodes/ have to go to the binutils list.

This patch needs a ChangeLog entry, documentation, and tests.

It has various formatting issues; I can nit-pick them all if you want,
but maybe after addressing the above would be better.

Francois> +  di.probe_address_func = (bfd_vma ( *) (char* , void* )) py_probe_address;

Casting function pointers is bad.

Francois> +  probe_func = PyObject_GetAttrString(gdb_module, "probe_address");
Francois> +  value = PyString_FromString(bufm);
Francois> +  result = PyObject_CallFunctionObjArgs (probe_func, value, NULL);

These things need error checks, and probably a call to gdbpy_print_stack
(or throwing a gdb exception).

This applies elsewhere.

Francois> +	bfd_vma probed_address = (*info->probe_address_func) (op_txt[i],
Francois> +	    info->application_data);

Am I right in thinking that the python function must parse the operand
to decide what to do?

If so, I wonder if something friendlier could be done.

Tom