On Thu, Nov 22, 2018 at 10:21:53AM +0100, Andreas Schwab wrote:
> On Nov 22 2018, Jakub Jelinek <ja...@redhat.com> wrote:
>
> > Is that with --enable-checking=release
>
> Yes.
Ok, reproduced myself, fixed thusly, committed to unbreak it.
cp_expr has
operator tree () const { return m_value; }
tree & operator* () { return m_value; }
tree operator* () const { return m_value; }
tree & operator-> () { return m_value; }
tree operator-> () const { return m_value; }
so it does the right thing most of the time, like with:
TREE_CHECK (NODE, something)->something_else
even if #define TREE_CHECK(NODE) NODE
But in this case we have:
#define USERDEF_LITERAL_VALUE(NODE) \
(((struct tree_userdef_literal *)USERDEF_LITERAL_CHECK (NODE))->value)
and so it works only if TREE_CHECK expands to something that forces the cast
to tree.
2018-11-22 Jakub Jelinek <ja...@redhat.com>
PR c++/87386
* parser.c (cp_parser_operator): Use str.get_value () instead of just
str in USERDEF_LITERAL_VALUE and USERDEF_LITERAL_SUFFIX_ID arguments.
--- gcc/cp/parser.c.jj 2018-11-21 23:40:10.999140630 +0100
+++ gcc/cp/parser.c 2018-11-22 10:19:47.384240264 +0100
@@ -15306,8 +15306,8 @@ cp_parser_operator (cp_parser* parser, l
return error_mark_node;
else if (TREE_CODE (str) == USERDEF_LITERAL)
{
- string_tree = USERDEF_LITERAL_VALUE (str);
- id = USERDEF_LITERAL_SUFFIX_ID (str);
+ string_tree = USERDEF_LITERAL_VALUE (str.get_value ());
+ id = USERDEF_LITERAL_SUFFIX_ID (str.get_value ());
end_loc = str.get_location ();
}
else
Jakub
