Evaluation of arguments in function calls

[Rajesh V]

Link to the question

20. main()
{
int i=5;
printf("%d%d%d%d%d%d",i++,i--,++i,--i,i);
}

The output is given as 45545. But I am getting 45555. Please clarify.

In general how to resolve these kind of problems?

[jorba101]

Link to the question

20. main()
{
int i=5;
printf("%d%d%d%d%d%d",i++,i--,++i,--i,i);
}

The output is given as 45545. But I am getting 45555. Please clarify.

In general how to resolve these kind of problems?

Nice problem.

I just realized compiling your example that parameters are read from right to left.

Anyway, I see logical the output 45545:

Take 5;

From right to left:

i : passes 5 to printf

--i: Substracts 1, and then prints. passes 4 to printf

++i: Adds 1, and then prints. passes 5 to printf

i--: passes 5 to printf, and then substracts one. We get 4

i++: passes 4 to printf, and then adds one. We get 5

Isn't it?

[JosAH]

Isn't it?

No it isn't; according to the Standard modifying a modifiable lvalue (such as
variable 'i') more than once before a sequence point has been reached causes
undefined behaviour and anything can happen.

kind regards,

Jos

[jorba101]

Does anybody know if is it reported / documented in GCC the described behavior?

[JosAH]

Does anybody know if is it reported / documented in GCC the described behavior?

There is no need to document that explicitly because the GCC compiler suite is
C99 conforming so that same piece of code produces undefined behaviour; just
as the Standard dictates. Of course a compiler produces 'something' but you
can't rely on it that another compiler will produce the same results.

kind regards,

Jos

[jorba101]

Of course a compiler produces 'something' but you

That's what I mean. I'm sure that GCC may produce C99-compliant code. And I'm not so sure that you can't configure it to do different things from the standard if you need too.

What I meant is that it is likely that a particular implementation like GCC produces an output which is following some kind of internal and reproduceable rule, which is also likely to be documented somewhere. Even though you can't rely another compiler will do different. But anyway, if documented, you can rely on what GCC will do.

[JosAH]

That's what I mean. I'm sure that GCC may produce C99-compliant code. And I'm not so sure that you can't configure it to do different things from the standard if you need too.

What I meant is that it is likely that a particular implementation like GCC produces an output which is following some kind of internal and reproduceable rule, which is also likely to be documented somewhere. Even though you can't rely another compiler will do different. But anyway, if documented, you can rely on what GCC will do.

That would be just stupid; never ever rely on what one particular compiler does.
If you ever need/want to port that code to another platform you're toast.

kind regards,

Jos

[Banfa]

In general all the problems on the page linked to produce undefined behaviour in some manor. I would recommend not trying to work through them.

[jorba101]

Hey,

That's ok. I wasn't gonna put that on production stuff.

I'm sure there's much more scaring things yet in the flashes of our TV's, DVD players or cars...

:D

[jorba101]

Anyway, following the topic on parameter evaluation order.

As per K&R2, section 2.12, parameter order evaluation is "unspecified" (in C99, in C89, or in both?, is C99 an ANSI?), so operators like "++i" should be better avoided if order is important.

Regarding modifying several lvalues at once in a statement, ok, i promise I won't do it (why couldn't just the compiler complain, to avoid these discussions?).


Btw, about GCC and C99 compliance:

http://gcc.gnu.org/gcc-4.3/c99status.html

[JosAH]

Regarding modifying several lvalues at once in a statement, ok, i promise I won't do it (why couldn't just the compiler complain, to avoid these discussions?).

Because the compiler can't always know or see it; ex:

/* in file A.c */
int foo(int* x) { return ++(*x); } 
 
 /* in file B.c */
extern int foo(int* x);
 
int main() {
 
   int i= 42;
   printf("i= %d\n, i+++foo(&i));
   return 0;
}
 

kind regards,

Jos

[Atos]

I don't think that C99 is a standard in the complete sense...
Most compilers ( if not all ), do not support it.

Anyway, this shuold not be the case however, cause you are talking for a non-portable code using a standard that was made to produce portable code!

As far as i know, function arguments are not guaranteed a standard order of evaluation.