Warnings with -Og - gcc 4.8

Warnings with -Og - gcc 4.8

[Lars Gullik Bjønnes]

I get a lot of warnings with -Og and gcc from trunk.
I do not get these warnings with any other warning level.

This is a reduced example:
----------
int foo(void);

int failing(void)
{
        int r;
        unsigned int i;
        struct h * p[3];

        for (i = 0; i < sizeof(p) / sizeof(p[0]); ++i) {
                r = foo();
                if (r != 0)
                        break;
        }
        return r >= 0;
}
-----------

Compile with: gcc -Og -Wall -c fil.c

gcc --version
gcc (GCC) 4.8.0 20130304 (experimental)

Should I create a bug on this, or will someone else handle that?

--
        Lgb

Re: Warnings with -Og - gcc 4.8

[Jeffrey Walton]

On Tue, Mar 5, 2013 at 1:41 PM, Lars Gullik Bjønnes <larsbj@gullik.org> wrote:
> I get a lot of warnings with -Og and gcc from trunk.
> I do not get these warnings with any other warning level.
>
> This is a reduced example:
> ----------
> int foo(void);
>
> int failing(void)
> {
>         int r;
>         unsigned int i;
>         struct h * p[3];
>
>         for (i = 0; i < sizeof(p) / sizeof(p[0]); ++i) {
>                 r = foo();
>                 if (r != 0)
>                         break;
>         }
>         return r >= 0;
> }
> -----------
>
> Compile with: gcc -Og -Wall -c fil.c
>
> gcc --version
> gcc (GCC) 4.8.0 20130304 (experimental)
>
> Should I create a bug on this, or will someone else handle that?
What warnings?

The code does not look correct to me (but I could be wrong). Does
'sizeof(p) / sizeof(p[0])' produce correct results when using
pointers?

Jeff

Re: Warnings with -Og - gcc 4.8

[Lars Gullik Bjønnes]

Jeffrey Walton <noloader@gmail.com> writes:

| On Tue, Mar 5, 2013 at 1:41 PM, Lars Gullik Bjønnes <larsbj@gullik.org> wrote:
>> I get a lot of warnings with -Og and gcc from trunk.
>> I do not get these warnings with any other warning level.
>>
>> This is a reduced example:
>> ----------
>> int foo(void);
>>
>> int failing(void)
>> {
>>         int r;
>>         unsigned int i;
>>         struct h * p[3];
>>
>>         for (i = 0; i < sizeof(p) / sizeof(p[0]); ++i) {
>>                 r = foo();
>>                 if (r != 0)
>>                         break;
>>         }
>>         return r >= 0;
>> }
>> -----------
>>
>> Compile with: gcc -Og -Wall -c fil.c
>>
>> gcc --version
>> gcc (GCC) 4.8.0 20130304 (experimental)
>>
>> Should I create a bug on this, or will someone else handle that?
| What warnings?

$ gcc -Og -Wall -c fil.c
fil.c: In function ‘failing’:
fil.c:14:11: warning: ‘r’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  return r >= 0;
           ^
| The code does not look correct to me (but I could be wrong). Does
| 'sizeof(p) / sizeof(p[0])' produce correct results when using
| pointers?

I belive that part to be correct.

-- 
	Lgb

Re: Warnings with -Og - gcc 4.8

[Jeff Law]

On 03/05/2013 12:04 PM, Lars Gullik Bjønnes wrote:
> Jeffrey Walton <noloader@gmail.com> writes:
>
> | On Tue, Mar 5, 2013 at 1:41 PM, Lars Gullik Bjønnes <larsbj@gullik.org> wrote:
>>> I get a lot of warnings with -Og and gcc from trunk.
>>> I do not get these warnings with any other warning level.
>>>
>>> This is a reduced example:
>>> ----------
>>> int foo(void);
>>>
>>> int failing(void)
>>> {
>>>          int r;
>>>          unsigned int i;
>>>          struct h * p[3];
>>>
>>>          for (i = 0; i < sizeof(p) / sizeof(p[0]); ++i) {
>>>                  r = foo();
>>>                  if (r != 0)
>>>                          break;
>>>          }
>>>          return r >= 0;
>>> }
>>> -----------
>>>
>>> Compile with: gcc -Og -Wall -c fil.c
>>>
>>> gcc --version
>>> gcc (GCC) 4.8.0 20130304 (experimental)
>>>
>>> Should I create a bug on this, or will someone else handle that?
> | What warnings?
>
> $ gcc -Og -Wall -c fil.c
> fil.c: In function ‘failing’:
> fil.c:14:11: warning: ‘r’ may be used uninitialized in this function [-Wmaybe-uninitialized]
>    return r >= 0;

Without the optimizer enabled, GCC does not do the analysis to prove the 
loop always executes and thus does not know it can eliminate the edge 
from the entry block to the return block with its associated 
uninitialized use.

jeff

Re: Warnings with -Og - gcc 4.8

[Lars Gullik Bjønnes]

Jeff Law <law@redhat.com> writes:

| On 03/05/2013 12:04 PM, Lars Gullik Bjønnes wrote:
>> Jeffrey Walton <noloader@gmail.com> writes:
>>
>> | On Tue, Mar 5, 2013 at 1:41 PM, Lars Gullik Bjønnes
>> | <larsbj@gullik.org> wrote:
>>>> I get a lot of warnings with -Og and gcc from trunk.
>>>> I do not get these warnings with any other warning level.
>>>>
>>>> This is a reduced example:
>>>> ----------
>>>> int foo(void);
>>>>
>>>> int failing(void)
>>>> {
>>>>          int r;
>>>>          unsigned int i;
>>>>          struct h * p[3];
>>>>
>>>>          for (i = 0; i < sizeof(p) / sizeof(p[0]); ++i) {
>>>>                  r = foo();
>>>>                  if (r != 0)
>>>>                          break;
>>>>          }
>>>>          return r >= 0;
>>>> }
>>>> -----------
>>>>
>>>> Compile with: gcc -Og -Wall -c fil.c
>>>>
>>>> gcc --version
>>>> gcc (GCC) 4.8.0 20130304 (experimental)
>>>>
>>>> Should I create a bug on this, or will someone else handle that?
>> | What warnings?
>>
>> $ gcc -Og -Wall -c fil.c
>> fil.c: In function ‘failing’:
>> fil.c:14:11: warning: ‘r’ may be used uninitialized in this function
>> [-Wmaybe-uninitialized]
>>    return r >= 0;
>
| Without the optimizer enabled, GCC does not do the analysis to prove
| the loop always executes and thus does not know it can eliminate the
| edge from the entry block to the return block with its associated
| uninitialized use.

I see that I have a mistake in my first post, I meant to say:

  I do not see this with any other _optimization_ level.

I do not get this warning with -O0, nor with -O1, only with -Og.

-- 
	Lgb

Re: Warnings with -Og - gcc 4.8

[Jeff Law]

On 03/05/2013 01:23 PM, Lars Gullik Bjønnes wrote:
>
>    I do not see this with any other _optimization_ level.
>
> I do not get this warning with -O0, nor with -O1, only with -Og.

A minimal amount of optimization is necessary to build the CFG & SSA 
graphs which are necessary for the uninitialized use analysis warning.

Thus with -O0 (do not optimize at all), you do not get any warning.

-Og means optimize, but only where doing so will not affect the quality 
of debug information.  So only a minimal set of optimizations are 
performed and you get the warning.

At higher optimization levels (-O1 and above) the compiler will prove 
the loop always executes and simplify the control flow graph and 
statements accordingly.  After those simplifications, the path 
containing the uninitialized use is eliminated (it's unreachable) and 
thus you do not get the warning.

It's unfortunate that this warning is not stable across different 
optimization levels, but for now that's simply how it works.

jeff

Re: Warnings with -Og - gcc 4.8

[Jeffrey Walton]

On Tue, Mar 5, 2013 at 3:39 PM, Jeff Law <law@redhat.com> wrote:
> On 03/05/2013 01:23 PM, Lars Gullik Bjønnes wrote:
>
>...
> -Og means optimize, but only where doing so will not affect the quality of
> debug information.  So only a minimal set of optimizations are performed and
> you get the warning.
For completeness and benefit to others, you still need to specify a
debugging information level (or take the default). See
http://gcc.gnu.org/ml/gcc/2013-02/msg00274.html for some details

OT: GCC 4.8 has some nice new features. -fsanitize=memory and
-fsanitize=threads look very interesting. Awesome work by the
developers.

Jeff