Re: [PATCH] Fix handling of DW_AT_data_bit_offset

[Tom Tromey <tromey@adacore.com>]

>>>>> "Tom" == Tom Tromey <tromey@adacore.com> writes:

Tom> A new test case is included.  This test fails with GCC 11, but passes
Tom> with an earlier version of GCC.

>> I'm checking this in now.

Tom> I spoke too soon - the new test fails now with the Fedora 34 system gcc.
Tom> I'll debug it a bit and see what sort of change is needed.

I finally looked.  The working DWARF has an array in a structure that
looks like:

 <3><11f9>: Abbrev Number: 10 (DW_TAG_array_type)
    <11fa>   DW_AT_name        : (indirect string, offset: 0x1a4d): pr__response__T2b
    <11fe>   DW_AT_type        : <0x1248>
    <1202>   DW_AT_sibling     : <0x1210>
 <4><1206>: Abbrev Number: 11 (DW_TAG_subrange_type)
    <1207>   DW_AT_type        : <0x1241>
    <120b>   DW_AT_upper_bound : <0x11a7>

... where the upper bound points at the first member of the enclosing
struct.

However, the failing DWARF does:

 <3><1058>: Abbrev Number: 11 (DW_TAG_array_type)
    <1059>   DW_AT_name        : (indirect string, offset: 0x1a87): pr__response__T2b
    <105d>   DW_AT_type        : <0x10a7>
    <1061>   DW_AT_sibling     : <0x106f>
 <4><1065>: Abbrev Number: 12 (DW_TAG_subrange_type)
    <1066>   DW_AT_type        : <0x10a0>
    <106a>   DW_AT_upper_bound : 3 byte block: 97 94 4 	(DW_OP_push_object_address; DW_OP_deref_size: 4)


Here, and I think correctly, gdb uses the address of the array in the
struct as the object address -- but gcc apparently expects the object
address to be the address of the enclosing structure.

This fix is in gcc 12, so I'm going to update the test case to xfail for
earlier versions of gcc, and then check it in.

Tom