__base_class_type_info offset_flags

[Mark Mitchell]

For the most part, the ABI specification doesn't offer particular
guidance for 32-bit platforms.  However, 2.9.5 (RTTI Layout) has this
note about the offset_flags field of __base_class_type_info:

    The upper 56 bits of __offset_flags are a signed offset. For a
    non-virtual base, this is the offset in the object of the base
    subobject. For a virtual base, this is the offset in the virtual table
    of the virtual base offset for the virtual base referenced (negative).

    The low-order byte of __offset_flags contains flags, as given by the
    masks from the enumeration __offset_flags_masks:

      o 0x1: Base class is virtual
      o 0x2: Base class is public

    Note:

    In an ABI for a machine with 32-bit addresses, the offset becomes a
    32-bit field, and the flags are in the low-order byte of a second
    32-bit field.

I think we should remove the note.

First, G++ does not honor the note; it just uses "long" on all
platforms.  That's a particular hard thing to change going forward;
the runtime support routines would have to somehow detect which
version of the type_info structures they were dealing with.  So, I
don't fancy trying to make this change in GCC.

Second, what's the point?  If you just deleted the note, you'd be
limited to objects whose sizes are representable in 24-bits on a
32-bit machine.  Well, more properly, if you made bigger objects,
you'd not be able to dynamic_cast between subobjects.

Any objections to simply removing the note and treating 32-bit
machines just like other machines, using a single "long" for
"__offset_flags"?

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