[Bug c++/11063] [3.3/3.4 Regression] ICE on parsing initialization list of const array member

[janis187 at us dot ibm dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11063


janis187 at us dot ibm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at redhat dot com


------- Additional Comments From janis187 at us dot ibm dot com  2003-08-05 22:45 -------
The regression (ICE) in PR 11063 was introduced or exposed by this patch:

--- gcc/gcc/cp/ChangeLog ---

2001-01-03  Jason Merrill  <jason@redhat.com>

        * init.c (build_default_init): New fn.
        (perform_member_init): Split out from here.
        (build_new_1): Use it.  Simplify initialization logic.
        (build_vec_init): Take an array, rather than a pointer and maxindex.
        Speed up simple initializations.  Don't clean up if we're assigning.
        * cp-tree.h: Adjust.
        * decl2.c (do_static_initialization): Remove TREE_VEC case.
        * parse.y (new_initializer): Return void_zero_node for ().
        * typeck.c (build_modify_expr): Handle getting a CONSTRUCTOR.
        * typeck2.c (digest_init): Only complain about user-written
        CONSTRUCTORs.

The regression hunt took place on i686-pc-linux-gnu using this test
case, included here because it's slightly different from the one in the
attachment:

class Foo
{
private:
  const int val_[2];

public:
  Foo(int, int);
};

Foo::Foo(int v0, int v1)
  : val_((int[]) {v0, v1})
{
}

Before this patch the test case got these messages:

11063.C: In constructor `Foo::Foo(int, int)':
11063.C:12: incompatible types in assignment of `int[2]' to `const int[2]'

With this test case I get the same ICE with a recent mainline compiler
built with and without checking disabled.