g++ bug? or am I missing something?

ezzetabi · 2008-08-19 14:02:08

#include <arpa/inet.h>

int main()
{
    uint32_t ip;
    ip = ::htonl(ip);
    return 0;
}

Saved as test.cc...

compiling...

$ g++ -g test.cc
$ g++ -O2 test.cc
test.cc: In function 'int main()':
test.cc:7: error: expected id-expression before '(' token
$ g++ --version
g++ (GCC) 4.3.1 20080724 (prerelease)
Copyright (C) 2008 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

The same code compiles with success or gives an error changing optimization?
What does it mean?

Last edited by ezzetabi (2008-08-19 14:03:32)

Kknd · 2008-08-20 02:37:49

If you remove the "::", it compiles with -O2. Don't know why!

Shaika-Dzari · 2008-08-20 02:51:14

Why did you write ::htonl ?
htonl is a simple function...

Allan · 2008-08-20 03:13:50

Shaika-Dzari wrote:

Why did you write ::htonl ?

That is just explicitly scoping the function to be from the global namespace. A bit overkill for this simple example but useful in more complicated code.

As far as I can tell, this code should compile and that it doesn't with -O2 is a bug. I am no C++ guru however...

Edit: gcc-3.3 may (NOT) help you figure this out!

test.cc: In function `int main()':
test.cc:6: error: syntax error before `__extension__'
test.cc:6: error: `__x' undeclared (first use this function)
test.cc:6: error: (Each undeclared identifier is reported only once for each 
   function it appears in.)
test.cc:6: error: `__v' undeclared (first use this function)
test.cc: At global scope:
test.cc:6: error: syntax error before `)' token

Allan · 2008-08-20 03:25:56

Running "gcc -E test.cc > raw.cc" and "gcc -O2 -E test.cc > opt.cc" and you will note what -O2 is doing to your code.

    ip = ::htonl(ip);

is being replaced with

    ip = ::(__extension__ ({ register unsigned int __v, __x = (ip); if (__builtin_constant_p (__x)) __v = ((((__x) & 0xff000000) >> 24) | (((__x) & 0x00ff0000) >> 8) | (((__x) & 0x0000ff00) << 8) | (((__x) & 0x000000ff) << 24)); else __asm__ ("rorw $8, %w0;" "rorl $16, %0;" "rorw $8, %w0" : "=r" (__v) : "0" (__x) : "cc"); __v; }));

Note the global namespace operator is still there which is wrong. So I say this is definitely a gcc bug.

ezzetabi · 2008-08-20 07:26:25

@Kknd
This is the reason of the whole topic... Yes?

@Shaika-Dzari
It is as Allan said. In C++ I always use ::f() for global or legacy C functions. It is about giving an immediate feedback to people who read the code that I am calling a global, not a member function.
Moreover I can make member functions with the same name of global ones without worrying.
Think about a socket class, it can well have a member function called bind(), yet there won't be ambiguity in the code if I always call ::bind() for the global function.
Of course in the example it is an overkill, but when you think you found a bug you have to make a minimal example to show the incorrect behaviour. And here it is.

@Allan
Programming 101: when in doubt use -E and see what the compiler is really seeing.
Yet, I did not thought -E -O2 was meaningful... I thought optimizations were made only after....
So, there is some kind of bug in function inlining. gcc inlines and forgets removing the scope operator.
Thanks a lot Allan.

Last edited by ezzetabi (2008-08-20 07:30:22)

Shaika-Dzari · 2008-08-20 14:22:26

Thank both of you. I have learning something useful today

piotroxp · 2008-08-30 11:52:41

int main()
{
    uint32_t ip;
    ip = (::htonl(ip));
    return 0;
}

Compiled perfectly, then tested w/ gdb for other stuff,

    .file    "test.cpp"
    .text
.globl main
    .type    main, @function
main:
.LFB2:
    leal    4(%esp), %ecx
.LCFI0:
    andl    $-16, %esp
    pushl    -4(%ecx)
.LCFI1:
    pushl    %ebp
.LCFI2:
    movl    %esp, %ebp
.LCFI3:
    pushl    %ecx
.LCFI4:
    subl    $20, %esp
.LCFI5:
    movl    -8(%ebp), %eax
    movl    %eax, (%esp)
    call    htonl
    movl    %eax, -8(%ebp)
    movl    $0, %eax
    addl    $20, %esp
    popl    %ecx
    popl    %ebp
    leal    -4(%ecx), %esp
    ret
.LFE2:
    .size    main, .-main
    .ident    "GCC: (GNU) 4.3.1 20080724 (prerelease)"
    .section    .note.GNU-stack,"",@progbits

<<disassembly.

No bug AFAIK. The scope operator tries to seek out the root of the obj/function, therefore using parentheses where you might use them gives the desired output in your case. When the parentheses are removed, it also does compile. No bug here.

Last edited by piotroxp (2008-08-30 14:00:56)

dante4d · 2008-09-13 22:46:16

I run into similar problem with Window$ API may times. Lots of "functions" are actually macros, where :: operator will fail.

robertp · 2008-09-14 09:19:50

The correct code is:

#include <arpa/inet.h>

int main()
{
    uint32_t ip = 0x01234567;
    ip = htonl(ip);
    return 0;
}

Using scope operator before htonl is invalid since it's a macro and macros don't have any scoping.

Arch Linux

#1 2008-08-19 14:02:08

g++ bug? or am I missing something?

#2 2008-08-20 02:37:49

Re: g++ bug? or am I missing something?

#3 2008-08-20 02:51:14

Re: g++ bug? or am I missing something?

#4 2008-08-20 03:13:50

Re: g++ bug? or am I missing something?

#5 2008-08-20 03:25:56

Re: g++ bug? or am I missing something?

#6 2008-08-20 07:26:25

Re: g++ bug? or am I missing something?

#7 2008-08-20 14:22:26

Re: g++ bug? or am I missing something?

#8 2008-08-30 11:52:41

Re: g++ bug? or am I missing something?

#9 2008-09-13 22:46:16

Re: g++ bug? or am I missing something?

#10 2008-09-14 09:19:50

Re: g++ bug? or am I missing something?

Board footer