On 02/13/2017 12:28 PM, Jakub Jelinek wrote:
The reason for that VOID_TYPE_P stuff in COND_EXPR handling is not that stabilize_expr doesn't like it, but to avoid warning twice. If I replace all 3 VOID_TYPE_P tests (with the patch) in cp_build_modify_expr with 0 && VOID_TYPE_P, the above testcase errors 6 times instead of 4 times as it should. For the case when lhs is a pre-inc/decrement or modify_expr, we don't have this problem, rhs is used just one in cp_convert_for_assignment where this error is diagnosed normally. Thus, I'll bootstrap/regtest following simplified version of the patch instead:
ok, thanks nathan -- Nathan Sidwell
