This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch+7.8] Fix 7.8 regression: resolve_dynamic_struct: Assertion `TYPE_NFIELDS (type) > 0' (PR 17642)


> It is another missing check_typedef().

> gdb/
> 2014-11-25  Jan Kratochvil  <jan.kratochvil@redhat.com>
> 
> 	Fix internal error on stubs of dynamic types.
> 	* gdbtypes.c (resolve_dynamic_type_internal): Apply check_typedef to
> 	TYPE if not TYPE_CODE_TYPEDEF.
> 
> gdb/testsuite/
> 2014-11-25  Jan Kratochvil  <jan.kratochvil@redhat.com>
> 
> 	Fix internal error on stubs of dynamic types.
> 	* gdb.base/vla-stub-define.c: New file.
> 	* gdb.base/vla-stub.c: New file.
> 	* gdb.base/vla-stub.exp: New file.

Pre-approved with the following comments. Ok for 7.8 as well, although
the chances of a 7.8.2 are getting slimmer, and you'll need also to
create a PR.

> diff --git a/gdb/gdbtypes.c b/gdb/gdbtypes.c
> index 61d259d..aa76604 100644
> --- a/gdb/gdbtypes.c
> +++ b/gdb/gdbtypes.c
> @@ -1875,41 +1875,48 @@ resolve_dynamic_type_internal (struct type *type, CORE_ADDR addr,
>    if (!is_dynamic_type_internal (real_type, top_level))
>      return type;
>  
> -  switch (TYPE_CODE (type))
> +  if (TYPE_CODE (type) == TYPE_CODE_TYPEDEF)
>      {
> -    case TYPE_CODE_TYPEDEF:
>        resolved_type = copy_type (type);
>        TYPE_TARGET_TYPE (resolved_type)
>  	= resolve_dynamic_type_internal (TYPE_TARGET_TYPE (type), addr,
>  					 top_level);
> -      break;
> +    }
> +  else 

Trailing space here...

> +    {
> +      /* Typedefs do not need to be preserved here but we need
> +	 to resolve any possible stub types.  */
> +      type = real_type;

It took me a moment, despite your added comment, to understand that
it was not so much about TYPE_CODE_TYPEDEF but about stubs! I keep
forgetting that check_typedef does more than unwrapping typedef
layers.

We know that type isn't a typedef here, so mentioning typedefs
is a bit confusing, IMO. How about we adjust the comment to say,
for instance:

        /* Before trying to resolve TYPE, make sure it is not a stub.  */

This is only a suggestion, and it's subjective, so feel free to ignore.

Thank you,
-- 
Joel


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]