You have to modify iCurArrayIdx after the statement, or perform the
assignment in the function call.
2009/4/27 Robin.Horde <robinhorde@xxxxxxxxxxxx>:
>
> Hi!
> I have a question about "global" variable accessing when gcc compile some code. See example below:
> //----------start-----------
> #include <stdio.h>
>
> int aiArray[10];
> int iCurArrayIdx;
>
> int GetNewValue( int iValue )
> {
> //do something...
>
> //i change the global variable intentionally.
> iCurArrayIdx = 1;
>
> //assume the return is 1
> return 1;
> }
>
> int main( int argc, char* argv[] )
> {
> aiArray[iCurArrayIdx] = GetNewValue( aiArray[iCurArrayIdx] );
>
> printf( "aiArray[iCurArrayIdx] = %d\n", aiArray[iCurArrayIdx] );
>
> return 0;
> }
> //----------end-------------
>
> I know the coding is very bad. It's simple for demo.
>
> I compile these code under VC6 and GCC-4.2.4/linux and run,
> then I get the differ result.
>
> VC6 printf: aiArray[iCurArrayIdx] = 1
> GCC printf: aiArray[iCurArrayIdx] = 0
>
> I think VC6 got the correct executable content.
>
> So I use objdump to get the assemble code of GCC output:
> 08048454 <_Z11GetNewValuei>:
> 8048454: 55 push %ebp
> 8048455: 89 e5 mov %esp,%ebp
> 8048457: c7 05 a8 97 04 08 01 movl $0x1,0x80497a8
> 804845e: 00 00 00
> 8048461: b8 01 00 00 00 mov $0x1,%eax
> 8048466: 5d pop %ebp
> 8048467: c3 ret
>
> 08048468 <main>:
> 8048468: 8d 4c 24 04 lea 0x4(%esp),%ecx
> 804846c: 83 e4 f0 and $0xfffffff0,%esp
> 804846f: ff 71 fc pushl -0x4(%ecx)
> 8048472: 55 push %ebp
> 8048473: 89 e5 mov %esp,%ebp
> 8048475: 53 push %ebx
> 8048476: 51 push %ecx
> 8048477: 83 ec 10 sub $0x10,%esp
> 804847a: 8b 1d a8 97 04 08 mov 0x80497a8,%ebx
> 8048480: a1 a8 97 04 08 mov 0x80497a8,%eax
> 8048485: 8b 04 85 80 97 04 08 mov 0x8049780(,%eax,4),%eax
> 804848c: 89 04 24 mov %eax,(%esp)
> 804848f: e8 c0 ff ff ff call 8048454 <_Z11GetNewValuei>
> 8048494: 89 04 9d 80 97 04 08 mov %eax,0x8049780(,%ebx,4)
> 804849b: a1 a8 97 04 08 mov 0x80497a8,%eax
> 80484a0: 8b 04 85 80 97 04 08 mov 0x8049780(,%eax,4),%eax
> 80484a7: 89 44 24 04 mov %eax,0x4(%esp)
> 80484ab: c7 04 24 90 85 04 08 movl $0x8048590,(%esp)
> 80484b2: e8 f5 fe ff ff call 80483ac <printf@plt>
> 80484b7: b8 00 00 00 00 mov $0x0,%eax
> 80484bc: 83 c4 10 add $0x10,%esp
> 80484bf: 59 pop %ecx
> 80484c0: 5b pop %ebx
> 80484c1: 5d pop %ebp
> 80484c2: 8d 61 fc lea -0x4(%ecx),%esp
> 80484c5: c3 ret
> 80484c6: 90 nop
> 80484c7: 90 nop
> 80484c8: 90 nop
> 80484c9: 90 nop
>
> Then I found the reason of run differ result:
> Address of 804847a, iCurArrayIdx has load in EBX
> Address of 8048494, use the old iCurArrayIdx keep in EBX and not update
>
> Puzzle! Is this GCC's bug, or feature, or standard ambiguity? Why be that?
>
> --horde
>
> ----------------------------------------------------------------------
> Ian reply:
> The order of evaluation of the left hand side and the right hand side of
> the assigment operator is unspecified in C. Therefore, the behaviour of
> this program is unspecified, because iCurArrayIdx could be retrieved
> before or after GetNewValue is called, and different compilers are free
> to implement this program differently.
>
> ----------------------------------------------------------------------
> Thanks Ian!
> Now can any people tell me how to ensure the order of evaluation to get the expectant result alike VC6. BTW:I wouldn't like to modify the expression:
> aiArray[iCurArrayIdx] = GetNewValue( aiArray[iCurArrayIdx] );
>
> --horde
>
>
>
>
>
>
>
> ___________________________________________________________
> 好玩贺卡等你发，邮箱贺卡全新上线！
> http://card.mail.cn.yahoo.com/
>
