Re: [PATCH] dwarf2: Fix dwarf stack fetch array view size mismatch

[Simon Marchi via Gdb-patches <gdb-patches@sourceware.org>]

On 8/8/22 07:24, Denis Lukianov wrote:
> Following change 4bce7cdaf4 "gdbsupport: add array_view copy function",
> dwarf stack fetch sometimes cause an internal-error in
> array_view::copy, where a gdb_assert expects the source and destination
> view sizes to match. When called from dwarf_expr_context::fetch_result
> sometimes the lengths don't match.
>
> Both the source and destination views each have a separate implicit
> length. The source is correctly sliced for the copy. However, the
> destination is passed with the full allocated length, which does not
> necessarily match the source length.
>
> This patch slices the destination to match the source length.
>
> diff --git a/gdb/dwarf2/expr.c b/gdb/dwarf2/expr.c
> index 3549745df04..aa203e87bfb 100644
> --- a/gdb/dwarf2/expr.c
> +++ b/gdb/dwarf2/expr.c
> @@ -1025,7 +1025,7 @@ dwarf_expr_context::fetch_result (struct type
> *type, struct type *subobj_type,
>               subobj_offset += n - max;
>
>             copy (value_contents_all (val).slice (subobj_offset, len),
> -                 value_contents_raw (retval));
> +                 value_contents_raw (retval).slice (0, len));
>           }
>           break;
>

Hi Denis,

Do you have a reproducer for this, that we could turn into a test?

I don't think your change is necessary. retval's content length comes
from the length of `subobj_type`:

  retval = allocate_value (subobj_type);

and the variable `len`, used to slice the source view, also comes from
the length of `subobj_type`:

  size_t len = TYPE_LENGTH (subobj_type);

So the two views should be of that same length.

Your patch reminded me of a pending patch I had in the same area, which
I just merged:

  https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=bde195b84a862f31ac111c0881ad13b89ee89492

Maybe you were seeing the same problem as described there?

Simon