Little C-function from the underworld
Recently, my student and I tried to understand one subtlety in the C standard. The easiest way to clarify such questions is to find out if the developers of the compilers took it into account, that is, to write code and see what different compilers will do with it.
I wrote this function:
Since
One suggestion was that the function would uniquely return " 0 ". Another is that the value of
This code only works on platforms where char is less than int, but I have not seen platforms where the opposite would have happened.
The first alarm bell rang when I compiled the code using Clang:
')
Not an ice at all - the function should in most cases return a unit. Then I realized that I was using an outdated version of Clang (2.7) and downloaded the last snapshot (rev 126534) (The article was written on March 3rd, 2011 - approx. Lane ):
Pay attention to the last sign. The result changed when we changed the optimization level - it’s normal if the value of the incremented CHAR_MAX is not defined, otherwise it is a compiler error.
Intel C compiler (12.0.2 for x86-64) behaved in a similar way:
The latest version of GCC (rev 170512 for x86) gives a stable result:
However, the story changes if we ask the compiler not to do function substitution:
Then I tried CompCert and the situation took an unexpected turn:
This is very strange, because CompCert contains a proven version of the implicit C language transformations - it was for their testing that
In short, since the type
Thus, the behavior is clearly defined and each correct compiler should produce the following result:
In other words, to calculate the result
So what about the fact that none of the four compilers that I usually use showed correct results?
Original article - blog.regehr.org/archives/482
I wrote this function:
int foo (char x) { char y = x; return ++x > y; } Since
++x increases the ++x value by 1, it is obvious that the function should return " 1 " for most x values. The question is what does it return for the value CHAR_MAX?One suggestion was that the function would uniquely return " 0 ". Another is that the value of
++x not defined on platforms where char is a signed type. At the end a test program was written, printing the conclusions for all possible options x : int main (void) { int i; for (i=CHAR_MIN; i<=CHAR_MAX; i++) { printf ("%d ", foo(i)); if ((i&31)==31) printf ("\n"); } return 0; } This code only works on platforms where char is less than int, but I have not seen platforms where the opposite would have happened.
The first alarm bell rang when I compiled the code using Clang:
')
regehr@home:~$ clang -O foo.c -o foo regehr@home:~$ ./foo 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Not an ice at all - the function should in most cases return a unit. Then I realized that I was using an outdated version of Clang (2.7) and downloaded the last snapshot (rev 126534) (The article was written on March 3rd, 2011 - approx. Lane ):
regehr@home:~$ clang -O0 overflow.c -o overflow regehr@home:~$ ./overflow 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 regehr@home:~$ clang -O1 overflow.c -o overflow regehr@home:~$ ./overflow 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Pay attention to the last sign. The result changed when we changed the optimization level - it’s normal if the value of the incremented CHAR_MAX is not defined, otherwise it is a compiler error.
Intel C compiler (12.0.2 for x86-64) behaved in a similar way:
[regehr@bethe ~]$ icc -O0 foo.c -o foo [regehr@bethe ~]$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 [regehr@bethe ~]$ icc -O foo.c -o foo [regehr@bethe ~]$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 The latest version of GCC (rev 170512 for x86) gives a stable result:
regehr@home:~$ current-gcc -O0 foo.c -o foo regehr@home:~$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 regehr@home:~$ current-gcc -O2 foo.c -o foo regehr@home:~$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 However, the story changes if we ask the compiler not to do function substitution:
regehr@home:~$ current-gcc -O2 -fno-inline foo.c -o foo regehr@home:~$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Then I tried CompCert and the situation took an unexpected turn:
regehr@home:~$ ccomp foo.c -o foo regehr@home:~$ ./foo 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 This is very strange, because CompCert contains a proven version of the implicit C language transformations - it was for their testing that
foo() was written.In short, since the type
char already of type int and when x is of CHAR_MAX type char and value CHAR_MAX , the behavior of ++x is clearly defined by both ANSI C and C99 standards.- The standard reads: “The expression
++Eequivalent to the expression(E+=1)”. - The standard says: "Using a compound assignment operator in the form
E1 op= E2differs from the usual assignmentE1 = E1 op (E2)only in that the variableE1calculated only once." - The right side of the assignment expression "
E1 op E2" refers to "ordinary arithmetic transformations". - Conventional arithmetic conversions imply that both operands to the "+" operator must be converted from the "
signed char" type to the "signed int" type before performing the addition operation. - Since int is wider than char, there is no likelihood of arithmetic overflow.
Thus, the behavior is clearly defined and each correct compiler should produce the following result:
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 In other words, to calculate the result
++x > y , when both variables are variables of the type " signed char " and have the value CHAR_MAX :- Signed char 127 is converted to signed int 127;
- Signed int 127 increases by one, turning into signed int 128;
- Signed int 128 is converted to signed char -128 — this is the new value of
xand the expressions++x(Note: this type of conversion is not defined by the standard, but is defined and must be clearly described by the specific implementation); - Signed char -128 is converted to signed int -128;
- A comparison of signed int -128> signed int 127 returns " 0 ".
So what about the fact that none of the four compilers that I usually use showed correct results?
- The gcc bug is known
- In LLVM / Clang, the bug was not known, but was fixed within 24 hours.
- The Intel compiler also contains an error.
- There is no error in CompCert. However, there is a problem of interaction between its definition of an unsigned char and the character values ​​CHAR_MIN and CHAR_MAX, described in the file limits.h on my linux machine. Verification of the compiler and its entire environment (header files, libraries, assemblers, I / O procedures, etc.) is a serious and open problem. That is why we write tests.
Original article - blog.regehr.org/archives/482
Source: https://habr.com/ru/post/147128/
All Articles