<style> .markdown-body code, code { color: #99999 !important; background-color: #ffe9e5; tab-size: 4; } </style> # 安全程式設計<br> assignment#1 110590004 林奕廷 --- ## 報告項目 + FLP34-C <br> [Ensure that floating-point conversions are within range of the new type](https://wiki.sei.cmu.edu/confluence/display/c/FLP34-C.+Ensure+that+floating-point+conversions+are+within+range+of+the+new+type) + FLP36-C <br> [Preserve precision when converting integral values to floating-point type](https://wiki.sei.cmu.edu/confluence/display/c/FLP36-C.+Preserve+precision+when+converting+integral+values+to+floating-point+type) --- ## C語言的型態轉換 顯性轉換與隱性轉換 ---- ```c= int A = 1; float B = 2.0f; double D = 4.0; A = B; // 隱性轉換, A = 2 A = int(D); // 顯性轉換, A = 4 A = sizeof(1 ? 1 : 0.9) // 隱性轉換, A = 8 (double) ``` ---- 三個基本隱性轉型規則 + 佔記憶體小 –> 佔記憶體大 + 整數 -> 浮點數 + 有號數 -> 無號數 --- ### FLP34-C Ensure that floating-point conversions are within range of the new type ---- ```c= // limits.h #define INT_MIN (-2147483647 - 1) #define INT_MAX 2147483647 // float.h #define DBL_MAX 1.7976931348623158e+308 #define DBL_MIN 2.2250738585072014e-308 #define FLT_MAX 3.402823466e+38F #define FLT_MIN 1.175494351e-38F ``` ```c= sizeof(int) // 4 byte sizeof(float) // 4 byte sizeof(double) // 8 byte ``` ---- ```c= float x = 1e5F; // 10^5 float y = 1e20F; // 10^20 double z = 1e300; // 10^300 int num_i; // range: -2*10^9 ~ 2*10^9 float num_f; // range: ±1*10^-38 ~ 3*10^38 num_i = x; // good! in int range num_i = 10000 num_i = y; // bad! not in int range num_i = ??? num_f = z; // bad! not in float range num_f = ??? ``` ---- how to resolve this? ---- ```c= #include <limits.h> #include <float.h> #include <math.h> int x; float y = ...; if (isgreater(fabsf(y), INT_MAX)) // handle undefined behavior else x = y; ``` ---- ```c= #include <limits.h> #include <float.h> #include <math.h> float y; double z = ... ; if (isgreater(fabs(z), FLT_MAX) || (z != 0.0 && isless(fabs(z), FLT_MIN) )) // handle error else y = z; ``` --- ### FLP36-C Preserve precision when converting integral values to floating-point type 將整數轉換為浮點數時保持精度 ---- 浮點數精度: IEEE754 標準下浮點數精度 float: 2^23 = 8388608 double:2^52 = 4503599627370496 ```c= int a = 1 << 10; // 2^10 = 1024 long long int b = 1LL << 40; // 2^40 int c = 1 << 24; // 2^24 long long int d = 1LL << 60; // 2^60 float num_f; double num_d; num_f = a; // 2^10 < 2^23 good! num_f = b; // 2^40 > 2^23 bad! num_d = c; // 2^24 < 2^52 good! num_d = d; // 2^60 > 2^52 bad! ``` ---- How to resolve this? ---- ```c= #incldue <stdlib.h> int x = ...; float y; if (abs(x) > (1 << 23)) // handle error else y = x; ``` ---- ```c= #incldue <stdlib.h> long long int x = ...; double y; if (llabs(x) > (1 << 52)) // handle error else y = x; ``` --- ## 結論 + 顯隱性轉型功能 + 潛在問題 + 謹慎處理 --- ## 報告結束