[aspectc-user] LNK2005 with woven Codes

[Olaf Spinczyk]

Hello Yan,

did you use *two* translation units, e.g. main.cc and my_class.cc? This 
is important. If you did, it looks as if you had to look deeper into the 
compilation and linking process of your build environment. If ac++ 
generates the same woven code within VisualStudio and in your 
environment, it is not an ac++ problem. I'm sorry, but I can't help you 
with these issues from here.

Best regards,

Olaf

Yan Mao wrote:
> Hello Olaf,
> sorry to bother you again, but I tested this inline function, it seems to work correctly.
> I define this global_counter function in a header file, and call it in a class memeber function and main method, and I get the correct answer, that the counter increments by each call.
> can you give some other tips?
> thanks,
> Yan
> 
> 
> 
> -------- Original-Nachricht --------
> Datum: Fri, 27 Oct 2006 10:41:50 +0200
> Von: Olaf Spinczyk <Olaf.Spinczyk at informatik.uni-erlangen.de>
> An: Yan Mao <maoyan at gmx.net>
> Betreff: Re: [aspectc-user] LNK2005 with woven Codes
> 
>> Hello Yan,
>>
>> let me explain COMDAT support with a simple code example:
>>
>> inline int global_counter () {
>>    static int c = 0;
>>    return c++;
>> }
>>
>> If this inline function is define in a *header* file, the compiler and 
>> linker have a problem, because the code is "seen" by the compiler in 
>> more than one translation unit. In this particular example this is no 
>> problem with the code for the function as it is inlined anyway. However, 
>> the function contains a static local variable, which means that some 
>> global storage has to be allocated. Although compiled in more than one 
>> translation unit, the compiler and linker have to make sure that this 
>> global storage is allocated only once! This is COMDAT support.
>>
>> Normally it is implemented by a concept known as "vague linkage", i.e. 
>> symbols in different translation units marked as "vague" (or "link 
>> once") are regarded as *one* by the linker if they have the same name 
>> and size.
>>
>> AspectC++ relies on this compiler/linker feature. This is normally no 
>> problem, but with your special build environment ...
>>
>> Maybe you could check this with the simple example shown above.
>>
>> Best regards,
>>
>> Olaf
>>
>>
>> Yan Mao wrote:
>>> Hallo Olaf,
>>>
>>> I did some experiments and I am now almost sure that, my tool works
>> different as VS Tool.
>>> I tested the same woven codes with both VS Tool and my special Tool,
>> they gave me too difference results. With My tool, Aspects are instantiated
>> twice, but with VS Tool only once.
>>> I posted the woven codes and different results, please take a look, and
>> maybe you can give me some advice, how to avoid the problem.
>>> And what is "COMDAT-support", how can I make sure, that if my tool has
>> "COMDAT-support" or not?
>>> Thanks,
>>>
>>> Yan
>>>
>>> p.s. Because both English and German are not my mother language, and I
>> am not sure that my German is much better than my English, so i still use
>> English.
>>>
>>>
>>> -------- Original-Nachricht --------
>>> Datum: Tue, 24 Oct 2006 05:08:12 +0200
>>> Von: Olaf Spinczyk <Olaf.Spinczyk at informatik.uni-erlangen.de>
>>> An: Yan Mao <maoyan at gmx.net>
>>> Betreff: Re: [aspectc-user] LNK2005 with woven Codes
>>>
>>>> Hello Yan,
>>>>
>>>> to be honest, your description of the problem is not very precise and I
>>>> have no chance to reproduce the misbehavior.
>>>>
>>>> If both advices are defined in the same aspect and if you don't define
>>>> your own aspect instantiation strategy with "aspectof()", I don't see a
>>>> reason why there should be two aspect instances. Or do you use some
>>>> special linker that has no "COMDAT-support"? Could you send me more
>>>> information about your build environment? If it helps, you could also
>>>> write in german.
>>>>
>>>> - Olaf
>>>>
>>>> Yan Mao wrote:
>>>>> Hallo Olaf,
>>>>>
>>>>> thanks for the answer. I changed my codes, and they worked now well. 
>>>>>
>>>>> But as I tried to use the same aspect in another project, I got
>> another
>>>> problem.  
>>>>>  
>>>>> In my aspect,I want to break an action, if the execution time expands.
>>>>>
>>>>> In an advice of one pointcut I set the "start time", and in the other
>>>> advice of another pointcut, i'd like to read the "start time", so that
>> i can
>>>> compare the "current time" with the "start time". 
>>>>> But unfortunately, i cant get the setted start time. 
>>>>>
>>>>> It seemes that, 2 Aspect object are created in der Laufzeit, where
>> only
>>>> one object should be created. That's why I cant get the set value.
>>>>> Actually the AC woven codes are almost the same with the sample
>> project,
>>>> which worked well. But They just don't work!
>>>>> To notice, that my current project are build on a system, which wraps
>>>> the Visual Studio APIs and Libraries, and offers its own APIs. And I
>> must
>>>> also use its own Compiler to compile and link the projects, which
>> should call
>>>> VS compiler cl.exe internally. 
>>>>> I suppose that it has somthing to with the compiler, but i dont know
>> how
>>>> to get the wright way to solve the problem.
>>>>> Can you give me some tipps?
>>>>> Sorry for my poor English, if it is hard to read. But I really need
>>>> help!
>>>>> Thanks in advance,
>>>>> Yan
>>>>>  
>>>>>
>>>>>
>>>>> -------- Original-Nachricht --------
>>>>> Datum: Sun, 22 Oct 2006 08:20:11 +0200
>>>>> Von: Olaf Spinczyk <Olaf.Spinczyk at informatik.uni-erlangen.de>
>>>>> An: Yan Mao <maoyan at gmx.net>
>>>>> Betreff: Re: [aspectc-user] LNK2005 with woven Codes
>>>>>
>>>>>   
>>>>>> Hello Yan,
>>>>>>
>>>>>> it looks as if you defined the member functions of "Profiling" as
>>>>>> non-inline functions in the aspect header file. If the aspect affects
>>>>>> more than one translation unit, the code will be generated more than
>>>>>> once and the linker complains. Better define the small functions as
>>>>>> inline functions.
>>>>>>
>>>>>> Best regards,
>>>>>>
>>>>>> Olaf
>>>>>>
>>>>>> Yan Mao wrote:
>>>>>>     
>>>>>>> Hello,
>>>>>>>
>>>>>>> I got LNK2005 errors by linking the woven codes.
>>>>>>> it seems to me, that the same aspect is expanded in multiple cpp
>>>> files.
>>>>>>> I have #ifndef #define and #endif macros in every head file, and I
>>>> also
>>>>>>>       
>>>>>> notice that AC++ compiler generate constant names  such as
>>>>>>     
>>>>>>> __ac_have_C__DA_ACSTUTest_aspects_Profiling_ah__
>>>>>>> __ac_need_C__DA_ACSTUTest_aspects_Profiling_ah__
>>>>>>>
>>>>>>> But they still dont work. What's wrong with my codes?
>>>>>>>
>>>>>>> I'd like to post the sample codes, hope anyone can help me!
>>>>>>>
>>>>>>> thanks in advance!!
>>>>>>>
>>>>>>> Yan
>>>>>>>   
>>>>>>>       
>>>>>   
>