Re: [C++ PATCH]: size_zero_node

[Mark Mitchell <mark at codesourcery dot com>]

Nathan --

  This patch looks pretty good.  It looks like I slightly misguessed
what that code in cp/typeck.c was for; I thought it was for handling
partial instantiations (which got a non-sizetype integer zero node as
their size as a marker that they were partial instantiations.)  But,
now I see what's going on.

  First, I think we need an better abstraction here.  Let's make a
COMPLETE_TYPE_P macro, where we can hide whatever complexity is
required.  There are all kinds of places in the C++ front-end where it
says `TYPE_SIZE (t) != NULL_TREE' and that's not nearly as intuitive
as saying `COMPLETE_TYPE_P'.  

  Also, why does `void' need a size at all?  Does this have to do with
making the GCC extension:

  void *p;
  p++;

work?  Otherwise, size `void' is an incomplete type, it would seem
that we could simply not set its TYPE_SIZE; no other incomplete types
have a TYPE_SIZE.  Perhaps this is set to simplify checking whether
something like:

  void f() {}

is legal; here, the return type must be a complete type, or void, even
though void is incomplete.  Perhaps by setting TYPE_SIZE for void, we
can get away with just checking TYPE_SIZE.  But, then, we have to
specifically check for `void' on:

  void v;

rather than just checking that the TYPE_SIZE is non-NULL.

  I suggest:

  o Don't set TYPE_SIZE for `void'.

  o Define COMPLETE_TYPE_P as macro which checks for non-NULL
    TYPE_SIZE.

  What do people think of this plan?

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com