[aspectc-user] volatile kills args (...) matching

[Olaf Spinczyk]

Hi,

I'd like to clarify something about the C++ standard here. The following 
function signatures are all considered as semantically equivalent in C++:

void f(int);
void f(const int);
void f(volatile int);
void f(const volatile int);

This is fact relevant for function overloading (which does not work with 
these four functions), linking, etc.

Therefore, every C++ compiler including the ac++ weaver normalizes these 
signatures. BTW, NOT ALL "const" and "volatile" keywords are removed, 
just those on the top level of an argument type. For example, the 
signature "void f(const int *)" remains as it is and is not considered 
equivalent with "void f(int *)".

Now it should be clear why an AspectC++ match expression "void f(int)" 
matches all four functions shown above.

The meaning of "const" and "volatile" in a match expression is 
different: It is considered as a restriction. For example, a match 
expression "void %(int *)" would match "void f (const int*)" and "void 
f(int*)", while "void %(const int*)" ONLY matches "void f(const int*). A 
match expression "void f(const %)" does not make sense and will yield no 
results, because after normalization no function still has a matching 
signature.

I don't want to say with this mail that the behavior of ac++ is correct 
and make sense. It's just to explain a few of the effects that you've 
observed.

Cheers,

Olaf


arnd-hendrik.mathias at nefkom.net wrote:
> Hi,
> I don't think it is really a bug but depends rather on the way, the 
> pointcut handling and the before(), after()...etc. functions work. My 
> test application shows following behavior:
>
> - volatile does not influence the matching of pointcuts to the 
> functions itself, which means that even
>   + a pointcut with a volatile parameter matches a function with 
> non-volatile parameter and
>   + a pointcut with a non-volatile parameter matches a function with 
> volatile parameters.
>
> - using volatile in the parameter list of before ( ... ) etc. causes 
> the advice to be silently ignored in any case (no warning, no error 
> but no working, too).
>
> - omitting volatile in the parameter list of the before ( ... ) etc. 
> still fits the parameter even of a volatile parameter pointcut.
>
> I try to attach the example code to this mail. If it turns out to be 
> filtered out I'll cite it in another one.
> Best regards
>
> Arnd-Hendrik
>
>
> Zitat von Panu Bloigu <panu.bloigu at gmail.com>:
>
>> Hello.
>>
>> arnd-hendrik.mathias at nefkom.net wrote:
>>> Hi,
>>> I just recognized one strange effect with parameter types in  
>>> pointcuts: If I declare one parameter of a function as volatile  
>>> type (in the source code as well as in the aspect) ac++ does no  
>>> longer match this execution (nor call) pointcut ( execution ( "%  
>>> blah ( ... )" ) && args ( bli, blubb )).
>>> Has anyone experienced this before?
>>> Regards
>> If you think that you have found a bug in ac++, then could you please
>> post a compilable code example that demonstrates the problem. If at all
>> possible, it would be good to try to minimize the code example so that
>> it still is compilable and still demonstrates the problem. If your
>> findings indeed turn out to be a bug in ac++, then I'm pretty sure the
>> AspectC++ developers will appreciate that you reported it.
>>>
>>>
>>> Arnd-Hendrik
>>>
>>> P.S.: I know that a volatile parameter in a function looks somewhat 
>>>  uncommon, but I just stripped it down for debugging purposes. In  
>>> real life it's a typedef'ed HW-Register value and this must be  
>>> volatile.
>>
>> It's hard to say if this is relevant in your case, but I think that
>> using type aliases (i.e. typedef'd type names) in pointcut descriptions
>> is considered problematic. I once posted a question about this on the
>> list, and here's Olaf's explanation on the issues involved:
>> http://aspectc.org/pipermail/aspectc-user/2007-January/001098.html
>>
>> -- Panu
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> aspectc-user mailing list
> aspectc-user at aspectc.org
> http://www.aspectc.org/mailman/listinfo/aspectc-user