This testcase was breaking because we found ourselves in operand_equal_p with a COND_EXPR whose operand 2 was null -- i.e., missing else branch, which won't happen for ? : but can happen here via -Wduplicated-branches.
Bootstrapped/regtested on x86_64-linux, ok for trunk? 2017-01-26 Marek Polacek <pola...@redhat.com> PR c/79199 * fold-const.c (operand_equal_p) [COND_EXPR]: Use OP_SAME_WITH_NULL for the third operand. * c-c++-common/Wduplicated-branches-13.c: New test. diff --git gcc/fold-const.c gcc/fold-const.c index 5576d59..a8bb8af 100644 --- gcc/fold-const.c +++ gcc/fold-const.c @@ -3147,7 +3147,7 @@ operand_equal_p (const_tree arg0, const_tree arg1, unsigned int flags) TREE_OPERAND (arg1, 0), flags)); case COND_EXPR: - if (! OP_SAME (1) || ! OP_SAME (2)) + if (! OP_SAME (1) || ! OP_SAME_WITH_NULL (2)) return 0; flags &= ~OEP_ADDRESS_OF; return OP_SAME (0); diff --git gcc/testsuite/c-c++-common/Wduplicated-branches-13.c gcc/testsuite/c-c++-common/Wduplicated-branches-13.c index e69de29..7aa5b37 100644 --- gcc/testsuite/c-c++-common/Wduplicated-branches-13.c +++ gcc/testsuite/c-c++-common/Wduplicated-branches-13.c @@ -0,0 +1,23 @@ +/* PR c/79199 */ +/* { dg-do compile } */ +/* { dg-options "-Wduplicated-branches" } */ + +unsigned int a, b, c, d, e; +void +fn1 (void) +{ + if (0) /* { dg-warning "this condition has identical branches" } */ + { + if (d > 4294967293) + (void) 5; + c = d; + b = e | a; + } + else + { + if (d > 4294967293) + (void) 5; + c = d; + b = e | a; + } +} Marek