Patch for bug with new-initializers

[Mark Mitchell <mmitchell at usa dot net>]

Here's a patch for code like this:

  void f()
  {
    new int = 1;
  }

This GNU extension didn't work at all.  Jason, may I check this in
with a testcase?

BTW, I am very dubious of GNU extensions (such as implicit typename,
these extended new-initializers, labeled initializers) and so forth,
because they seem to cause quite a few bugs, are not tested often
enough, and add complexity when trying to work on the C++ front-end.
They also create noise on the list; when deciding whether to fix a
problem or not, I have to figure out whether it's one I care about,
and I generally don't care about these extensions.  However, I do want
the compiler to be crash-proof on *all* inputs, even invalid ones.
Life would be simpler if these crashes simply resulted in error
messages when the extension was attempted; then users would better
understand what is going on.

I suggest disabling some of these (perhaps the set I mentioned above)
by default; an error message is better than a crash.  They could be
enabled by individual switches by interested parties.  What do y'all
think of this idea?

-- 
Mark Mitchell <mmitchell@usa.net>
http://home.earthlink.net/~mbmitchell
Consulting Services Available

Tue Mar 10 23:23:20 1998  Mark Mitchell  <mmitchell@usa.net>

	* parse.y (new_initializer): Make sure all initializers are
	lists.

Index: parse.y
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/cp/parse.y,v
retrieving revision 1.39
diff -c -p -r1.39 parse.y
*** parse.y	1998/03/02 12:22:31	1.39
--- parse.y	1998/03/11 07:30:44
*************** new_initializer:
*** 1217,1223 ****
  		{
  		  if (pedantic)
  		    pedwarn ("ANSI C++ forbids initialization of new expression with `='");
! 		  $$ = $2;
  		}
  	;
  
--- 1217,1226 ----
  		{
  		  if (pedantic)
  		    pedwarn ("ANSI C++ forbids initialization of new expression with `='");
! 		  if (TREE_CODE ($2) != TREE_LIST)
! 		    $$ = build_expr_list (NULL_TREE, $2);
! 		  else
! 		    $$ = $2;
  		}
  	;