Even a >= check would have been suboptimal. Rather than /* generated code */ if ( ++p == pe ) goto _test_eof; or /* generated code */

Even a >= check would have been suboptimal. Rather than /* generated code */ if ( ++p == pe ) goto _test_eof; or /* generated code */ if ( ++p >= pe ) goto _test_eof; they should have had /* generated code */ if ( ++p == pe ) goto _test_eof; assert(p < pe); since having servers core dumping would have drawn attention to the bug in a way that counting one byte too many and then hitting _test_eof would not.