Hi,
this rather old issue points out that, for the snippet below in the
legacy -std=c++98 and -pedantic, we issue an hard error instead of a
pedwarn - what we normally do, as clarified by Manuel. I guess we could
imagine further clean-ups in this area but something as simple as the
below seems correct to me and pretty conservative.
Tested x86_64-linux.
Thanks, Paolo.
///////////////////////
/cp
2016-11-17 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/55080
* parser.c (cp_parser_primary_expression): Issue a pedwarn instead
of an error floating-point literals in a constant-expression.
/testsuite
2016-11-17 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/55080
* g++.dg/parse/pr55080.C: New.
Index: cp/parser.c
===================================================================
--- cp/parser.c (revision 242540)
+++ cp/parser.c (working copy)
@@ -4840,8 +4840,10 @@ cp_parser_primary_expression (cp_parser *parser,
cast is to an integral or enumeration type will be
checked at that point. If we are not within a cast, then
this code is invalid. */
- if (!cast_p)
- cp_parser_non_integral_constant_expression (parser, NIC_FLOAT);
+ if (!cast_p && !parser->allow_non_integral_constant_expression_p)
+ pedwarn (input_location, OPT_Wpedantic,
+ "ISO C++ forbids using a floating-point literal in a "
+ "constant-expression");
}
return cp_expr (token->u.value, token->location);
Index: testsuite/g++.dg/parse/pr55080.C
===================================================================
--- testsuite/g++.dg/parse/pr55080.C (revision 0)
+++ testsuite/g++.dg/parse/pr55080.C (working copy)
@@ -0,0 +1,6 @@
+// PR c++/55080
+// { dg-options "-std=c++98 -pedantic" }
+
+class B {
+ static const int c = 3.1415926; // { dg-warning "constant-expression" }
+};
