http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37237
--- Comment #11 from Tom Tromey <tromey at gcc dot gnu.org> 2013-02-18 15:20:56
UTC ---
(In reply to comment #10)
> I don't think such an attribute belongs in the DWARF standard, since this is
> very much an internal detail of the ABI; another ABI might have just a single
> destructor with magic arguments, like the old G++ ABI.
>
> You need to know a lot about the destructor ABI as it is; looking for D0/D1
> doesn't seem like a major additional burden to me.
With git master I actually see multiple destructors now:
<1><be>: Abbrev Number: 15 (DW_TAG_subprogram)
<bf> DW_AT_abstract_origin: <0x93>
<c3> DW_AT_linkage_name: (indirect string, offset: 0x10): _ZN1AD2Ev
<c7> DW_AT_low_pc : 0x0
<cf> DW_AT_high_pc : 0x39 0x0
<d7> DW_AT_frame_base : 1 byte block: 9c (DW_OP_call_frame_cfa)
<d9> DW_AT_object_pointer: <0xe1>
<dd> DW_AT_GNU_all_tail_call_sites: 1
<dd> DW_AT_sibling : <0xea>
[...]
<1><ea>: Abbrev Number: 17 (DW_TAG_subprogram)
<eb> DW_AT_abstract_origin: <0x93>
<ef> DW_AT_linkage_name: (indirect string, offset: 0x3e): _ZN1AD0Ev
<f3> DW_AT_low_pc : 0x3a
<fb> DW_AT_high_pc : 0x26 0x0
<103> DW_AT_frame_base : 1 byte block: 9c (DW_OP_call_frame_cfa)
<105> DW_AT_object_pointer: <0x109>
<109> DW_AT_GNU_all_tail_call_sites: 1
What I'd like to know is what is guaranteed.
Previously gcc didn't emit the linkage name for any destructor -- but
this would make the proposed solution much harder to implement (gdb
would have to implement name mangling...).
So I suppose I'd like a guarantee that the destructor will be emitted
with at least one linkage name.
