Re: Enum initialization modifies expression's type

[Jeffrey A Law <law at cygnus dot com>]

  In message <orhfboejl4.fsf@tamanduatei.dcc.unicamp.br>you write:
  > --=-=-=
  > 
  > The testcase in the attached patch failed because build_enumerator()
  > modified the type of the initializer, that was a constant obtained
  > from the size_table, and already referenced as TYPE_SIZE of the
  > `void*' type.  Then, when the size was given to size_binop(), it
  > didn't have the same type as the other argument, so we crashed.
  > 
  > This patch fixes this problem.  Ok to install?
  > 
  > 
  > --=-=-=
  > Content-Type: text/x-patch
  > Content-Disposition: inline; filename=enum-sizeof.patch
  > 
  > Index: gcc/ChangeLog
  > from  Alexandre Oliva  <aoliva@cygnus.com>
  > 
  > 	* c-decl.c (build_enumerator): Don't modify the value's type,
  > 	convert it.
I was going to recommend you install a similar patch for C++, but looking
at its build_enumerator function, I think it already handles this case
by making a copy of the node in question (if its necessary).

Can you try the code in the C++ front-end and see if it fixes the
problem?  If so, I would prefer to use the same solution in the C
front-end as we're using in the C++ front-end.

I'm assuming the problem is clobbering the TREE_TYPE of VALUE, right?

The C++ code in question which I think handles this issue:

  /* We always have to copy here; not all INTEGER_CSTs are unshared.
     Even in other cases, we will later (in finish_enum) be setting
     the type of VALUE.  But, we don't need to make a copy if this
     VALUE is one of the enumeration constants for this same
     enumeration type.  */
  for (values = TYPE_VALUES (enumtype); values; values = TREE_CHAIN (values))
    if (TREE_VALUE (values) == value)
      break;
  /* If we didn't break out of the loop, then we do need a copy.  */
  if (!values && value)
    value = copy_node (value);

  > 	* gcc.c-torture/compile/20000523-1.c: New test.
This is fine.  Please install the new test.

jeff