On Thu, Nov 30, 2017 at 1:26 PM, David Malcolm <dmalc...@redhat.com> wrote: > +/* Test if EXP is merely a wrapper node, added to express a location_t > + on behalf of the node's child (e.g. by maybe_wrap_with_location). > + > + A wrapper node has code NON_LVALUE_EXPR or VIEW_CONVERT_EXPR, and the > + same type as its operand. > + > + NON_LVALUE_EXPR is used for wrapping constants. > + VIEW_CONVERT_EXPR is used for wrapping non-constants. > + > + A subtlety is that we may have to test whether we have the correct > + TREE_CODE for the wrapped TREE_CODE. Otherwise, e.g. the C++ expression: > + reinterpret_cast<int>(some_int_var) > + is a NON_LVALUE_EXPR around a non-constant of the same type, and > + could thus be mischaracterized as a location wrapper node. > + > + Hence we need to check CONSTANT_CLASS_P (TREE_OPERAND (EXP, 0)) > + for a NON_LVALUE_EXPR for EXP to be a wrapper. > + We don't need to check a VIEW_CONVERT_EXPR that wraps the same type > + for !CONSTANT_CLASS_P; such instances of VIEW_CONVERT_EXPR are always > + wrappers. */
I think most of this comment (everything after the first paragraph) belongs inside the function rather than before it. OK with that change. Jason
