On Wed, Oct 13, 2010 at 2:10 PM, Jan Kratochvil
<jan.kratochvil@redhat.com> wrote:
> Also I do not like much the pr[0-9]+.exp names, one gets fimiliar with any
> name but I never remember PR numbers.  [in fact an idea by Roland McGrath]

I often just go with the flow for things such as this, in the absence
of specific rules.
If we want to declare a specific rule of no more such prNNN naming,
that's fine by me.

> @@ -1388,7 +1389,7 @@ check_typedef (struct type *type)
>          /* It is dangerous to call lookup_symbol if we are currently
>             reading a symtab.  Infinite recursion is one danger.  */
>          if (currently_reading_symtab)
> -           return type;
> +           return make_cv_type (is_const, is_volatile, type, NULL);
>
>          name = type_name_no_tag (type);
>          /* FIXME: shouldn't we separately check the TYPE_NAME and

I left this out for a later pass to be conservative and avoid having
to deal with any unintended consequences.
But if one wants to handle this in the same patch, ok.

> @@ -1398,7 +1399,7 @@ check_typedef (struct type *type)
>          if (name == NULL)
>            {
>              stub_noname_complaint ();
> -             return type;
> +             return make_cv_type (is_const, is_volatile, type, NULL);
>            }
>          sym = lookup_symbol (name, 0, STRUCT_DOMAIN, 0);
>          if (sym)

Ditto.

> @@ -1407,11 +1408,10 @@ check_typedef (struct type *type)
>            TYPE_TARGET_TYPE (type) = alloc_type_arch (get_type_arch (type));
>        }
>       type = TYPE_TARGET_TYPE (type);
> +      is_const |= TYPE_CONST (type);
> +      is_volatile |= TYPE_VOLATILE (type);
>     }
>
> -  is_const = TYPE_CONST (type);
> -  is_volatile = TYPE_VOLATILE (type);
> -
>   /* If this is a struct/class/union with no fields, then check
>      whether a full definition exists somewhere else.  This is for
>      systems where a type definition with no fields is issued for such

This part doesn't feel right.
For example, when a stub is later filled in the accumulated cv
qualifiers will get filled in too, but that's wrong.
[Right?]

Plus there's more than just const/volatile to preserve.
[Right?]

I like this patch better.


2010-10-13  Doug Evans  <dje@google.com>
            Jan Kratochvil  <jan.kratochvil@redhat.com>

        PR exp/12117
        * gdbtypes.c (check_typedef): Clean up function comment.
        Keep track of instance flags as we strip typedefs and create a new
        type to preserve them if necessary.

        testsuite/
        * gdb.cp/ptype-cv-cp.cc: New file.
        * gdb.cp/ptype-cv-cp.exp: New file.