Re: [PATCH 3/4] gdb: Migrate frame unwinders to use C++ classes

[Lancelot SIX <lancelot.six@amd.com>]

On Wed, Mar 06, 2024 at 01:51:34PM +0100, Guinevere Larsen wrote:

Hi Gwen,

I have not read the specific of this series, but testing the patches as
is causes a build failure on amdgpu-tdep.c

> diff --git a/gdb/amdgpu-tdep.c b/gdb/amdgpu-tdep.c
> index 8189de8e565..45c83a562e8 100644
> --- a/gdb/amdgpu-tdep.c
> +++ b/gdb/amdgpu-tdep.c
> @@ -892,7 +892,7 @@ amdgpu_frame_prev_register (const frame_info_ptr &this_frame, void **this_cache,
>    return frame_unwind_got_register (this_frame, regnum, regnum);
>  }
>  
> -static const frame_unwind amdgpu_frame_unwind = {
> +static const frame_unwind_legacy amdgpu_frame_unwind (
>    "amdgpu",
>    NORMAL_FRAME,
>    FRAME_UNWIND_ARCH,
> @@ -903,7 +903,7 @@ static const frame_unwind amdgpu_frame_unwind = {
>    default_frame_sniffer,
>    nullptr,
>    nullptr,
            ^
Moving from the aggregate initializer to a regular constructor call,
this last coma must be removed.

With this fixed, I tested this series on the gdb.rocm/*.exp part (both
on upstream GDB and our downstream port) and see no regression.

I'll try to actually read the actual later.

Best,
Lancelot.

> -};
> +);
>  
>  static int
>  print_insn_amdgpu (bfd_vma memaddr, struct disassemble_info *info)
> 
> -- 
> 2.43.2
>