Hi! On the following testcase we ICE, because op1 doesn't have normal int type, but a distinct attribute type, while their main variants are different, they are still treated as compatible types and the C++ FE doesn't actually try to ensure both arguments have the same type.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2018-11-28 Jakub Jelinek <ja...@redhat.com> PR c++/88215 * c-ubsan.c: Include langhooks.h. (ubsan_instrument_division): Change gcc_assert that main variants of op0 and op1 types are equal to gcc_checking_assert that the main variants are compatible types. * c-c++-common/ubsan/pr88215.c: New test. --- gcc/c-family/c-ubsan.c.jj 2018-11-14 17:44:02.159912988 +0100 +++ gcc/c-family/c-ubsan.c 2018-11-27 12:30:15.211687948 +0100 @@ -31,6 +31,7 @@ along with GCC; see the file COPYING3. #include "stringpool.h" #include "attribs.h" #include "asan.h" +#include "langhooks.h" /* Instrument division by zero and INT_MIN / -1. If not instrumenting, return NULL_TREE. */ @@ -44,8 +45,9 @@ ubsan_instrument_division (location_t lo /* At this point both operands should have the same type, because they are already converted to RESULT_TYPE. Use TYPE_MAIN_VARIANT since typedefs can confuse us. */ - gcc_assert (TYPE_MAIN_VARIANT (TREE_TYPE (op0)) - == TYPE_MAIN_VARIANT (TREE_TYPE (op1))); + tree top0 = TYPE_MAIN_VARIANT (type); + tree top1 = TYPE_MAIN_VARIANT (TREE_TYPE (op1)); + gcc_checking_assert (lang_hooks.types_compatible_p (top0, top1)); op0 = unshare_expr (op0); op1 = unshare_expr (op1); --- gcc/testsuite/c-c++-common/ubsan/pr88215.c.jj 2018-11-27 12:33:59.102998653 +0100 +++ gcc/testsuite/c-c++-common/ubsan/pr88215.c 2018-11-27 12:33:35.694384374 +0100 @@ -0,0 +1,11 @@ +/* PR c++/88215 */ +/* { dg-do compile } */ +/* { dg-options "-fsanitize=integer-divide-by-zero" } */ + +int +foo (void) +{ + int a = 2, __attribute__ ((__unused__)) b = 1; + int f = a / b; + return f; +} Jakub