gimplify_parameters

[Richard Kenner]

    I thought that gimplify_type_sizes didn't actually modify the type node
    itself, that all the modification was reserved for the SAVE_EXPR within.

    That's clearly not the case now, but I think that is the Correct solution
    to the bug.

Hmm... it would seem right to me to do this in the gimplify_parameters
case, but less right in the "normal" case (that of a DECL_EXPR).  Do we
really want to *require* the SAVE_EXPR there?

I doubt this will work, but I'd have to run a lot of code through it to
see if that's the case.