Re: [PATCH] Convert dw2-const.exp to use the DWARF assembler

[Tom de Vries <tdevries@suse.de>]

On 5/29/26 4:58 PM, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:
> 
> Tom> Then I ran it with target board unix/-m32, and ran into:
> Tom> ...
> Tom> (gdb) print/x val8^M
> Tom> $2 = 0x0^M
> Tom> (gdb) FAIL: gdb.dwarf2/dw2-const.exp: print/x val8
> Tom> ...
> 
> Yeah, me too.
> 
> I wonder if there's some gas oddity here.  The DWARF doesn't seem to be
> notably weird or word-size-dependent.  I see in the old .S:
> 
>>> -#if 0 /* gas encodes it explicitly.  */
>>> -        .slab128        0xffffffff00000000
>>> -#endif
>>> -#if 0 /* gas encoded it correctly - formed GDB FAILed.  */
>>> -        .sleb128        -4294967296
>>> -#endif
>>> -#if 1 /* Rather encode sleb128 by hand.  */
>>> -        .byte           0x80, 0x80, 0x80, 0x80, 0x70
>>> -#endif
> 
> So hmm.
> 
> But maybe this is some kind of extension we would want to support in the
> DWARF assembler.
> 
> Anyway still looking into this.

I found the reason for the discrepancy.

Before the patch, with unix/m32, the adress size is 8:
...
   Compilation Unit @ offset 0x2df:
    Length:        0x6f (32-bit)
    Version:       2
    Abbrev Offset: 0x1bb
    Pointer Size:  8
  <0><2ea>: Abbrev Number: 1 (DW_TAG_compile_unit)
     <2eb>   DW_AT_stmt_list   : 0x200
     <2ef>   DW_AT_name        : file1.txt
     <2f9>   DW_AT_producer    : GNU C 3.3.3
     <305>   DW_AT_language    : 1       (ANSI C)
...

After the patch, it's 4.

In dwarf_expr_context::execute_stack_op, we do:
...
	case DW_OP_consts:
           op_ptr = safe_read_sleb128 (op_ptr, op_end, &offset);
           result = offset;
           result_val = value_from_ulongest (address_type, result);
           break;
...
and if address_type has size 4, result_val is 0.

Thanks,
- Tom