SIMPLE and fat pointers

[Richard Kenner]

    I had actually expected some GNAT people to join this discussion.
    After all, they will have a problem when they implement a pass to
    build functions as SIMPLE trees. I looked at the change logs, and
    _ALL_(*) changelog entries that involve these PLACEHOLDER_EXPR and 
    WITH_RECORD_EXPR nodes are from Richard Kenner, so I guess the only
    reason why these nodes are in the back end now is to support Ada...

    So I hope to hear from Richard Kenner ;-)

I don't understand the issue, but have not been following the SIMPLE
stuff much at all.

Certainly PLACEHOLDER_EXPR is required for Ada and not just for
unconstrained arrays: the more common use is in discriminated records.

The bounds of an array and position of a field will often be quite
complex expressions involving PLACEHOLDER_EXPR.  They cannot be simplified
until RTL generation since they represent different values for each
occurrence of the type.