Re: [RFA] Fix handling of "set sysroot foo"

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Fri, 12 Apr 2013 19:59:16 +0200, Doug Evans wrote:
> I kinda like and dislike clear_solist as the new name for free_so_symbols.
> - the old free_so_symbols does more than just free
> - target_so_ops.clear_solib is used for similar "reset and/or free" purposes
> - it's nice to have the target function's name match
> - target_so_ops function free_so already exists
> Hence the rename.

I would find more appropriate clear_so (for the function, the virtual method
declaration and its svr4 implementation).
 * clear_solib is for the whole backend, not just for one struct so_list.
 * free_so is for one struct so_list.

(There remains the problem "struct so_list" is used for both the entry and the
whole list as it contains "next".)

Also svr4_free_so could now have:
	gdb_assert (!so->lm_info->l_addr_p);


> --- solist.h	1 Jan 2013 06:32:51 -0000	1.38
> +++ solist.h	12 Apr 2013 17:56:09 -0000
> @@ -88,6 +88,10 @@ struct target_so_ops
       /* Free the link map info and any other private data structures
>         associated with a so_list entry.  */

+         clear_so method is always called before free_so is called.
+         SO is then never passed to the backend again.


>      void (*free_so) (struct so_list *so);
>  
> +    /* Reset or free private data structures associated with
> +       so_list entries.  */

Singular form.
    /* Reset or free private data structures associated with
       the SO entry.  */
+      SO then can be passed to the backend for futher use.


> +    void (*clear_solist) (struct so_list *so);
> +
>      /* Reset or free private data structures not associated with
>         so_list entries.  */
>      void (*clear_solib) (void);


Thanks,
Jan