C++ PATHC: PR5571

[Richard Kenner]

    > I'm not sure exactly what you are suggesting, though.  If you have to
    > call make_decl_rtl for a VAR_DECL when its type become complete, that
    > implies you have some way of finding all decls for that type.  Do you?

    In the C++ front end, we do.  There's no reason that, in any front end,
    you couldn't keep track of what variables you've created with incomplete
    types.

Well, the C front end doesn't, though I agree it *could*.

    Exactly my point.  The solution currently in expand_expr is designed
    to help you do things automatically -- the middle end fixes up the RTL
    for your variable for you.  But, that depends on you having completed
    the type already, which, for arrays, means that you have to have gone
    back and done that.  And if you can find the type, you can probably
    find the declaration, so you might as well fix that up too.

Yes and no.  I think the basic question is whether we want "incomplete
type" to be totally a front-end concept or have the middle end deal
with it as well.

    Having the middle end call layout_decl is indicative of a failure to
    adhere to that strategy: the middle end is now modifying trees.

I agree with this.