Re: const qualifiers in parse tree

[John W. Eaton]

On 12/08/2016 05:40 PM, Rik wrote:

> However, in some cases I ran in to problems with 'const'ness.  In
> pt-misc.cc, I made the following change:
>
> +verbatim+
> @@ -222,12 +215,8 @@ tree_parameter_list::variable_names (voi
>  {
>    std::list<std::string> retval;
>
> -  for (const_iterator p = begin (); p != end (); p++)
> -    {
> -      tree_decl_elt *elt = *p;
> -
> -      retval.push_back (elt->name ());
> -    }
> +  for (tree_decl_elt* elt : *this)
> +    retval.push_back (elt->name ());
>
>    return retval;
> -verbatim-
>
> However, this loses the notion of the 'const' property.  I know that
> getting the name() isn't going to change anything.  Should we be stricter
> in pt-decl.h and mark the name method as const?

Yes, if possible.  The parse tree code is old and I wasn't originally 
very careful with const.  So there's room for improvement there.  I've 
tried at various times to fix this, but ran into problems where trying 
to add const to one method caused a cascade of other changes that ran 
into a dead end of some function that I couldn't figure out how to make 
const.

> Also, I ran in to cases
> where I needed the identifier in a const manner.  Should we be declaring
> const/non-const versions of the ident method?
>
> +verbatim+
>   tree_identifier *ident (void) { return id; }
>
>   std::string name (void) { return id ? id->name () : ""; }
> -verbatim-
>
> Maybe this goes to
>
> +verbatim+
>   tree_identifier *ident (void) { return id; }
>   const tree_identifier *ident (void) { return id; }

Did you mean to have a const qualifier for the second function above?

>   std::string name (void) const { return id ? id->name () : ""; }
> -verbatim-

Yes, if this is enough to fix the kind of problem that I was describing 
above, then I'd say do it.  But maybe also add a note that it would be 
good to eliminate the non-const version at some point if possible.

jwe