PATCH: RTL_EXPR vs. inlining-on-trees

[Richard Kenner]

      This removal will have a very large cost for Ada.

    That's a valuable consideration.

Indeed now that I think about it, the issue isn't the *cost*, but that
it will *break* Ada!

The problem is this: I'm expanding some complex expression that
involves multiple temporaries (none of which survive past the code)
and I'm going to put it someplace else in the code.  I need those
temporaries to survive until such time as I expand the RTL_EXPR.  The
code that was there guarantees that.  Now that you've removed it, how
can that be done?