Re: [patch] Fix GC issue triggered by arithmetic overflow checking

[Richard Biener]

On Thu, Oct 13, 2016 at 12:15 PM, Eric Botcazou <ebotca...@adacore.com> wrote:
>> Yes.  But that's not the only source for DECL_UID differences.  Btw,
>> I see lots of FOR_EACH_HASH_TABLE_ELEMENT in var-tracking.c
>> but they don't look like their outcome is supposed to be dependent on
>> element ordering.
>
> This leads to NOTE_INSN_VAR_LOCATION notes emitted in a different order, which
> are then interpreted by dwarf2out_var_location.  In particular:
>
> (note 6350 6349 6351 (var_location temp (nil)) NOTE_INSN_VAR_LOCATION)
> (note 6351 6350 6352 (var_location temp$low (mem/c:DI (plus:SI (reg/f:SI 30
> %fp)
>         (const_int -112 [0xffffffffffffff90])) [10 MEM[(struct cpp_num
> *)&result + 8B]+0 S8 A64])) NOTE_INSN_VAR_LOCATION)
> (note 6352 6351 6353 (var_location temp$8 (nil)) NOTE_INSN_VAR_LOCATION)
> [...]
> (code_label 2091 6355 2092 79 912 "" [1 uses])
> (note 2092 2091 5271 79 [bb 79] NOTE_INSN_BASIC_BLOCK)
>
> is interpreted differently from:
>
> (note 6350 6349 6351 (var_location temp (nil)) NOTE_INSN_VAR_LOCATION)
> (note 6351 6350 6352 (var_location temp$8 (nil)) NOTE_INSN_VAR_LOCATION)
> (note 6352 6351 6353 (var_location temp$low (mem/c:DI (plus:SI (reg/f:SI 30
> %fp)
>         (const_int -112 [0xffffffffffffff90])) [10 MEM[(struct cpp_num
> *)&result + 8B]+0 S8 A64])) NOTE_INSN_VAR_LOCATION)
> [...]
> (note 2092 2091 5271 79 [bb 79] NOTE_INSN_BASIC_BLOCK)
>
> @@ -32608,6 +32608,17 @@
>         .uleb128 0x8
>         .byte   0x93    ! DW_OP_piece
>         .uleb128 0x8
> +       .uaword .LLVL592-.LLtext0       ! Location list begin address
> (*.LLLST153)
> +       .uaword .LLVL597-.LLtext0       ! Location list end address
> (*.LLLST153)
> +       .uahalf 0x9     ! Location expression size
> +       .byte   0x93    ! DW_OP_piece
> +       .uleb128 0x8
> +       .byte   0x8e    ! DW_OP_breg30
> +       .sleb128 -112
> +       .byte   0x93    ! DW_OP_piece
> +       .uleb128 0x8
> +       .byte   0x93    ! DW_OP_piece
> +       .uleb128 0x8
>         .uaword .LLVL695-.LLtext0       ! Location list begin address
> (*.LLLST153)
>         .uaword .LLVL696-.LLtext0       ! Location list end address
> (*.LLLST153)
>         .uahalf 0xe     ! Location expression size
>
> probably because the non-null location comes last in the second case.

Definitely looks like a bug to me.  Can you open a PR for this so it doesn't get
lost?

Thanks,
Richard.

> --
> Eric Botcazou