Re: [Boost-users] [multi_index] SIGABRT in multi_index_container constructor

From: Manuel Jung (gzahl_at_[hidden])
Date: 2007-02-15 17:16:39

"JOAQUIN LOPEZ MU?Z" wrote:

> ----- Mensaje original -----
> De: Manuel Jung <gzahl_at_[hidden]>
> Fecha: Jueves, Febrero 15, 2007 9:01 pm
> Asunto: Re: [Boost-users] [multi_index] SIGABRT in
> multi_index_containerconstructor
> Para: boost-users_at_[hidden]
> 
>> Ok now i get close to solving the problem.
>> 
>> My original class is the following:
>> 
>> class queue : public queue_set
>> {
>> public:
>>    queue(){};
>>    virtual ~queue(){};
>>    void AddQuery(query::pointer Query);
>>    void RmQuery(query::pointer Query);
>>    void Update(query::pointer Query);
>>    boost::mutex m_Queue;
>>    typedef boost::shared_ptr<queue> pointer;
>> };
> [...]
>> If i use it in the example jung.cpp it works. My real world
>> program works
>> without these 4 lines:
>>    void AddQuery(query::pointer Query);
>>    void RmQuery(query::pointer Query);
>>    void Update(query::pointer Query);
>>    boost::mutex m_Queue;
>> 
>> but not with them enabled!
>> 
>> Ok after playing a little bit around with them, i can say the
>> following:
>> Enabling just one of these procedures in my real program -> i get
>> the crash.
>> No function, everything fine.
>> Whats going on here??? These are not called in the constructor!?
>> Up to now they are never ever called!
>> (But i need them of course ;-))
>> So whats going on, im frightend..
> 
> Well, I might be totally wrong, but the more you describe
> the problem the more it looks to me like some neighboring variable
> corrupting the space occupied by your queue object. I've got
> several questions and suggestions to try to test whether
> this hypothesis is true:
> 
> 1. The crashing queue object, is it a gobal variable? If so,
> can you try moving its definition so as to displace it
> with respect to other global variables in the same translation
> unit? I mean, if your code looks like:

Its not a global variable, but part of a class. Its a public object created
in a shared_ptr.

>   some_type1 var1;
>   some_type2 var2;
>   queue      q;
>   some_type3 var3;
> 
> try shuffling it a bit like vg:
> 
>   queue      q;
>   some_type1 var1;
>   some_type2 var2;
>   some_type3 var3;
> 
> You get the idea.
> 
> 2. [assuming the object is global] If there are no neighboring
> vars, try planting some safe areas around the queue object:
> 
>   char pre[256];
>   queue q;
>   char post[256];
> 
> 3. [assuming the object is global] Try, just for testing
> purposes, to momentarily comment the offending queue out
> and declaring a queue object as a stack variable:
> 
>   int main(){
>     queue q;
>     ...
>   }
> 
> Crash/no crash?

im tired and will try this later. Probably Saturday, cause i can test it
tommorow, but dont post it.. im in a train ;-).

> 4. Last of all, your problem involves an internal mutex
> object which is only defined when you set Boost.MultiIndex
> safe mode on (you've got it set on, right?) If you turn
> this mode off you won't get the crash, but this is IMHO
> just masking the actual problem, which could appear
> under a different guise later on.

yes, the safe mode is on with this code:
#if !defined(NDEBUG)
#define BOOST_MULTI_INDEX_ENABLE_INVARIANT_CHECKING
#define BOOST_MULTI_INDEX_ENABLE_SAFE_MODE
#endif

when i switched it off, the error vanished, but even better some
libcurl/curlpp related error vanished! I have even some strange error about
clearing the Multihandle of curlpp. He says he cannot free a pointer. I
dont know if these errors are really related... but probably they are? This
error popped up when exiting from the program, but i couldnt test it
really, cause i couldnt send a Strg-C when running in gdb... but it seemed
to exiting cleanly.

So i'll try variable shifting tommorow.
Goot night,
Manuel Jung

> Good luck, please keep me informed,
> 
> Joaquín M López Muñoz
> Telefónica, Investigación y Desarrollo