refinements to definition of TREE_READONLY ?

[Eric Botcazou]

> But it also tests TREE_STATIC which should not be true for automatic
> variables. The binds_local_p check is only to guard against overriding the
> symbol with an external definition.

You're right, TREE_STATIC is the key here, in other words you're entitled to 
assume that !DECL_INITIAL => DECL_INITIAL == 0 for TREE_READONLY+TREE_STATIC 
and not TREE_READONLY alone.

-- 
Eric Botcazou