Re: [2/3] RFC: merge symbol "ops" and "aclass" fields

[Jan Kratochvil <jan.kratochvil@redhat.com>]

Hi Tom,

such a minor sanity check, not sure if worth it:


On Wed, 16 Jan 2013 17:08:32 +0100, Tom Tromey wrote:
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -584,6 +587,30 @@ struct symbol_register_ops
>    int (*register_number) (struct symbol *symbol, struct gdbarch *gdbarch);
>  };
>  
> +/* Objects of this type are used to find the address class and the
> +   various computed ops vectors of a symbol.  */
> +
> +struct symbol_impl
> +{
> +  enum address_class aclass;
> +
> +  /* Used with LOC_COMPUTED.  */
> +  const struct symbol_computed_ops *ops_computed;
> +
> +  /* Used with LOC_REGISTER and LOC_REGPARM_ADDR.  */
> +  const struct symbol_register_ops *ops_register;
> +
> +  /* Used to store a per-symbol-reader datum.  */
> +  const void *reader_datum;

These three fields could be union.


> +};
[...]
> @@ -679,10 +692,22 @@ struct symbol
>  #define SYMBOL_TYPE(symbol)		(symbol)->type
>  #define SYMBOL_LINE(symbol)		(symbol)->line
>  #define SYMBOL_SYMTAB(symbol)		(symbol)->symtab
> -#define SYMBOL_COMPUTED_OPS(symbol)     (symbol)->ops.ops_computed
> -#define SYMBOL_REGISTER_OPS(symbol)     (symbol)->ops.ops_register
> +#define SYMBOL_COMPUTED_OPS(symbol) \
> +  ((symbol_impls[(symbol)->aclass_index]).ops_computed)
> +#define SYMBOL_REGISTER_OPS(symbol) \
> +  ((symbol_impls[(symbol)->aclass_index]).ops_register)
> +#define SYMBOL_READER_DATUM(symbol) \
> +  ((symbol_impls[(symbol)->aclass_index]).reader_datum)
>  #define SYMBOL_LOCATION_BATON(symbol)   (symbol)->aux_value


And these accessors could gdb_assert impl->ACLASS.  Currently if ACLASS does
not match it returns NULL, which would (correctly) crash on dereferencing
a method but such coding bug would not be noticed during accesses like:
  if (SYMBOL_COMPUTED_OPS (sym) == &dwarf2_loclist_funcs)


Thanks,
Jan