TREE_READONLY, DECL_INITIAL, and gimplification

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

Are we heading in the same direction with it as we are with RTX_UNCHANGING_P
in that we allow *exactly one* assignment? 

That caused all sorts of semantic confusion with RTX_UNCHANGING_P and
I'm concerned we're now doing the same with TREE_READONLY.

In GENERIC, there's no confusion because the initial value is in
DECL_INITAL, so there's no assignment to a TREE_READONLY decl.  But
when we gimplify, we remove it from there and generate the assignment
to the TREE_REAONLY value, just like in the RTL case.  So there is
such an assignment in GIMPLE.

Is this likely to cause as much trouble as it did for RTL?