[PATCH v2] c++: Implement -Wuninitialized for mem-initializers (redux) [PR19808]

[Marek Polacek]

On Thu, Nov 18, 2021 at 05:10:47PM -0500, Jason Merrill wrote:
> On 11/8/21 18:41, Marek Polacek wrote:
> > @@ -1311,13 +1462,25 @@ emit_mem_initializers (tree mem_inits)
> >     if (!COMPLETE_TYPE_P (current_class_type))
> >       return;
> > +  /* Keep a set holding fields that are not initialized.  */
> > +  hash_set<tree> uninitialized;
> > +
> > +  /* Initially that is all of them.  */
> > +  if (warn_uninitialized)
> > +    for (tree f = next_initializable_field (TYPE_FIELDS (current_class_type));
> > +	 f != NULL_TREE;
> > +	 f = next_initializable_field (DECL_CHAIN (f)))
> > +      if (!DECL_ARTIFICIAL (f))
> > +	uninitialized.add (f);
> 
> I wonder about flipping the sense of the set, so that it tracks fields that
> have been initialized rather than those that haven't; then you wouldn't need
> this loop.

True, but then I'd have to figure out a new way to signal that we don't want
to warn about the current member-initializer-list.  What I mean by that is
that when I see e.g. a MODIFY_EXPR or something else with side-effects in
a mem-init, I can just empty the set:

    case MODIFY_EXPR:
    /* Don't attempt to handle statement-expressions, either.  */
    case STATEMENT_LIST:
      uninitialized->empty ();

and then we won't even bother walking the other mem-inits, because
find_uninit_fields has:

  if (!uninitialized->is_empty ())
    {
      // walk_tree ()
    }

and I thought that was pretty elegant.  Of course, I could just add a new
bool member into find_uninit_data and then do what you suggest...  Up to
you, I'm happy to do that too.

Marek