PR preprocessor/55715 libcpp: * expr.c (num_binary_op): Implement subtraction directly rather than with negation and falling through into addition case. gcc/testsuite: * gcc.dg/cpp/expr-overflow-1.c: New test. From-SVN: r205846