Uninitialized variables & Bounds runtime checking

[Miguel Angel Ajo]

Hi Axel, thank you very much!,

    I think that checking would need us to rewrite all of our class
definitions to
include this type of initialization.

   In the other hand, I know valgrind, and I use it regularly, but I
cannot use it on embedded platforms, as it really takes a lot of memory.

   I was looking for something like Mic--soft compilers have.. keeping
uninitialized values with a known constant, and when the
attribute/variable is accessed, and
this constant is detected, a runtime exception is thrown...  I know
it's not a perfectmethod, but it's quite useful :-)




2010/10/6 Axel Freyn <axel-freyn@gmx.de>:
> Hi Miguel,
> On Wed, Oct 06, 2010 at 09:23:39AM +0200, Miguel Angel wrote:
>>    I've been searching about possible *run-time* checks that can be
>> used with GCC, and after a little bit of investigation I discovered
>> about -fbounds-check, but no others...
>>
>>   -fbounds-check only apply to Fortran compiles or they work on C/C++ too?
>>
>> does exist some kind of flag for testing on unitialized variables
>> (during runtime), in my case some problems arise when somebody leaves
>> an unintialized C++ class attribute.
>
> Maybe you can use "-Weffc++" -- this warns e.g. for un-initialized
> attributes -- however,
>  - it generates MANY warnings, also from code in the standard library
>  - it will warn also when you assign a value in the constructor, but not
>   using initialization. The code:
>    class A{
>      A():x(0){}
>      A(const A&a){x=a.x;}
>      int x;
>    };
>   generates a warning for line 3: "warning: ‘A::x’ should be
>   initialized in the member initialization list"
>
> And of course, that's only a compile time check.
>
> I don't know any additional flags for gcc -- however: Do you know
> "valgrind"? That's an independent program, which can be used to perform
> checks for using of uninitialized memory on already compiled code -- it
> will report the corresponding line in the source-code too.
>
> Axel
>