Hi Simon,

Ok, I will describe my point of view with details...

For now we have this implementation (from the CVS):

1642 void H323ExtendedVideoCapability::AddAllCapabilities(
1643 H323Capabilities & basecapabilities, PINDEX descriptorNum,PINDEX simultaneous)
1644 {
1645 H323ExtendedVideoFactory::KeyList_T extCaps = H323ExtendedVideoFactory::GetKeyList();
1646 if (extCaps.size() > 0) {
1647 H323CodecExtendedVideoCapability * capability = new H323CodecExtendedVideoCapability();
1648 H323ExtendedVideoFactory::KeyList_T::const_iterator r;
1649 PINDEX num = P_MAX_INDEX;
1650 for (r = extCaps.begin(); r != extCaps.end(); ++r) {
1651 H323Capability * childCap = H323ExtendedVideoFactory::CreateInstance(*r);
1652 if (childCap != NULL) {
1653 H323CodecExtendedVideoCapability * extCapability = (H323CodecExtendedVideoCapability *)capability->Clone();
1654 extCapability->AddCapability(childCap);
1655 num = basecapabilities.SetCapability(descriptorNum, simultaneous, extCapability);
1656 simultaneous = num;
1657 delete childCap;
1658 }
1659 }
1660 simultaneous = P_MAX_INDEX;
1661 basecapabilities.SetCapability(descriptorNum, simultaneous,new H323ControlExtendedVideoCapability());
1662 delete capability;
1663 }
1664 }
2023 void H323CodecExtendedVideoCapability::AddCapability(const PString & cap)
2024 {
2025 extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap));
2026 }
2027
2028 void H323CodecExtendedVideoCapability::AddCapability(H323Capability * capability)
2029 {
2030 extCapabilities.Add((H323Capability *)capability->Clone());
2031 }


I see some issues for current implementation:

1.If someone just calls H323CodecExtendedVideoCapability::AddCapability(const PString & cap) manually - we will have the crash in result because of "double delete" for same pointer. The is absolutelly same issue from my first email. And that issue still not solved.
2.We have new AddCapability(H323Capability*) finction which has potential (and real) memory leak because it's behaviour a bit strange and not transparent as for me. This function confused me. For example, we have H323Capability * H323Capability::Create(const PString & name) function and this function doing that that it should.
3.Fix for H323ExtendedVideoCapability::AddAllCapabilities function doing some redundant work - basically all that new work it is deal of AddCapability(const PString&) function that we have had before the fix. Also, checking to !NULL has no sense, because it works with internal KeyList that returns the Factory, that factory has real pairs <keyname,pointer_to_object> and it is impossible to get a Key without an Object... We have no memory leak but it is simplest to use AddCapability(const PString&) with simplest fix for same goal.
My proposition the same I have had from the beginning:
1.Remove H323CodecExtendedVideoCapability::AddCapability(H323Capability * capability) function - we don't need it at all. If someone need to have this function - don't need to clone the pointer like we have now. It is only confuse us, plus  we have a potential memory leak in result.
2.Revert H323ExtendedVideoCapability::AddAllCapabilities back as we have had before the fix - first version of that function is absolutelly clear and has no any problem.
3.Modify H323CodecExtendedVideoCapability::AddCapability(const PString & cap) function to "clone" the object that returns by H323ExtendedVideoFactory::CreateInstance(cap) code - in resul H323CodecExtendedVideoCapability object will work with copy object only and will delete copy object too. No any memory leak or other undefined behavior.

Also possible to modify a bit H323CodecExtendedVideoCapability::AddCapability(const PString & cap) function, but it is not mandatory... I'm talking about a case when someone puts wrong parameter, for example a string which has no related Capability - for this case the Factory returns NULL and we will have a crash during attempt to clone the NULL-capability.
I can propose to replace that func to something like this (in result we able to control wrong parameters):

H323Capability* H323CodecExtendedVideoCapability::AddCapability(const PString & cap)
{
    H323Capability* p = H323ExtendedVideoFactory::CreateInstance(cap);
    if (!p)
      return p;

    p = (H323Capability*)p->clone();
    extCapabilities.Add(p);

    return p;
}
Thank's and good luck
--
Iurii Gordiienko

24 January 2016, 23:22:24, by "Simon Horne" <s.horne@packetizer.com>:

Iurii

 

I’m confused. There is no difference in the functionality except to resolve the memory leak. H323ExtendedVideoFactory::CreateInstance may return a NULL (unlikely tho’) which also needs to be checked. The change simply separates the CreateInstance, checks for NULL then clone() in basically the same place (another function) as the old implementation.

 

You are correct in that VLD like Valgrind detects a leak with the codec factory system with plugin codecs because the codecs are loaded as pointers to the singleton. The singleton is only created at startup and destroyed at shutdown so do not grow on a call-by-call basis.

 

There is also a small leak per call with some MediaOptions not being cleaned up. This is an issue as this is on a call-by-call basis. Fixes for that are welcome.

 

Simon

 

From: Iurii Gordiienko [mailto:hordi@ukr.net]
Sent: Monday, January 25, 2016 4:27 AM
To: Simon Horne <s.horne@packetizer.com>
Cc: 'Jan Willamowius' <jan@willamowius.de>; h323plus@lists.packetizer.com
Subject: Re[2]: [h323plus] Double-delete Capability bug

 

I'm sure - current implementations ::AddCapability are wrong - we should't clone the pointer for H323CodecExtendedVideoCapability::AddCapability(H323Capability * capability) function and should clone for H323CodecExtendedVideoCapability::AddCapability(const PString & cap) function. I think, Valgrind shows memory leak for case when we are using extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap)->clone()) (forr H323CodecExtendedVideoCapability::AddCapability(const PString & cap)) because we have no any default mechanism to clear (delete all capabilities) capabilities factory - we have static factory objects as I remember and Valgrind don't see any desctructor execution. I will write (tomorrow ) simple app and show you proper behavior and destructors execution.

We can't use current version (with clone()) of H323CodecExtendedVideoCapability::AddCapability(H323Capability * capability) fuction because that function has been declared as owner of pointer to H323Capability and any user expects that H323CodecExtendedVideoCapability object will delete that pointer in own destructor. Whis current modiification we have 100% memory leak because it has not declared behaviour.

 

 

2023

void H323CodecExtendedVideoCapability::AddCapability(const PString & cap)

2024

{

2025

    extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap));

2026

}

2027

 

2028

void H323CodecExtendedVideoCapability::AddCapability(H323Capability * capability)

2029

{

2030

    extCapabilities.Add((H323Capability *)capability->Clone());

2031

}

 

--- Original message ---
From: "Simon Horne" <s.horne@packetizer.com>
Date: 24 January 2016, 07:23:27

 

Fix checked in. 
The first instance is created in
H323ExtendedVideoCapability::AddAllCapabilities and a new function
H323CodecExtendedVideoCapability::AddCapability(H323Capability *) then
clones the first instance. The first instance is then deleted. This will
ensure every subsequent call to H323ExtendedVideoFactory::CreateInstance()
will create a new unique instance and there is no memory leak.
          
 

        
Simon
          
 

        
-----Original Message-----
From: h323plus [mailto:h323plus-bounces@lists.packetizer.com] On Behalf Of
Jan Willamowius
Sent: Sunday, January 24, 2016 2:35 AM
To: h323plus@lists.packetizer.com
Subject: Re: [h323plus] Double-delete Capability bug
          
 

        
Hi Simon,
          
 

        
it does need a better fix. But a leak is still better than a crash.
          
 

        
The proper fix is probably to have CreateInstance() always allocate new
memory. But until thats fixed I would leave the Clone() in.
          
 

        
Regards,
Jan
          
 

        
Simon Horne wrote:
> Jan
> 
> As I suspected, I just check with VLD on windows and the change causes 
> a 9kB/call memory leak (as the capability gets cloned 3 times per 
> call). Needs to be reverted out.
> 
> Simon
> 
> -----Original Message-----
> From: h323plus [mailto:h323plus-bounces@lists.packetizer.com] On 
> Behalf Of Jan Willamowius
> Sent: Saturday, January 23, 2016 4:49 AM
> To: h323plus@lists.packetizer.com
> Subject: Re: [h323plus] Double-delete Capability bug
> 
> Hi Iurii,
> 
> I have added the Clone() in the CVS, but could you please check (wg.
> with Valgrind) if we now have a memory leak in some cases instead of a 
> crash ?
> 
> Regards,
> Jan
> 
> --
> Jan Willamowius, Founder of the GNU Gatekeeper Project EMail  :
> jan@willamowius.de
> Website: http://www.gnugk.org
> Support: http://www.willamowius.com/gnugk-support.html
> 
> Relaxed Communications GmbH
> Frahmredder 91
> 22393 Hamburg
> Geschäftsführer: Jan Willamowius
> HRB 125261 (Amtsgericht Hamburg)
> USt-IdNr: DE286003584
> 
> Iurii Gordiienko wrote:
> > Hi,
> > I have found something... For
> H323ExtendedVideoCapability::AddAllCapabilities function we are using 
> H323CodecExtendedVideoCapability::AddCapability function for
> extCapability->AddCapability(*r).
> H323CodecExtendedVideoCapability::AddCapability function uses
> extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap))
function.
> The function H323ExtendedVideoFactory::CreateInstance(cap) will call "new"
> if any same "cap" does not is exists for now< for other case it will 
> return the pointer to current actual Capability. But 
> H323CodecExtendedVideoCapability::extCapabilities will delete all 
> items in time of execution  extCapabilities.RemoveAll() function from 
> H323CodecExtendedVideoCapability::~H323CodecExtendedVideoCapability().
> In result we have "double delete" and undefined behavior (crash for 
> worst case).
> > 
> > --------------------------------
> > void H323ExtendedVideoCapability::AddAllCapabilities(
> H323Capabilities & basecapabilities, PINDEX descriptorNum,PINDEX
> simultaneous) {   H323ExtendedVideoFactory::KeyList_T extCaps = 
> H323ExtendedVideoFactory::GetKeyList();   if (extCaps.size() > 0) { 
> H323CodecExtendedVideoCapability * capability = new 
> H323CodecExtendedVideoCapability();
> H323ExtendedVideoFactory::KeyList_T::const_iterator r;         PINDEX 
> num = P_MAX_INDEX;         for (r = extCaps.begin(); r != 
> extCaps.end(); ++r) {
>          H323CodecExtendedVideoCapability * extCapability = 
> (H323CodecExtendedVideoCapability *)capability->Clone();
>  extCapability->AddCapability(*r);            num = 
> basecapabilities.SetCapability(descriptorNum, 
> simultaneous,extCapability);
>          simultaneous = num;         }     simultaneous = P_MAX_INDEX; 
> basecapabilities.SetCapability(descriptorNum, simultaneous,new 
> H323ControlExtendedVideoCapability());  !
> >     delete capability;   }  }
> > void H323CodecExtendedVideoCapability::AddCapability(const PString &
> > cap) {
> > extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap)); 
> > }
> > 
> > -------------------------------------------------
> > I propose to replace the "void
> H323CodecExtendedVideoCapability::AddCapability(const PString & cap)" 
> to this one:
> > void H323CodecExtendedVideoCapability::AddCapability(const PString &
> > cap) {
> > extCapabilities.Add(H323ExtendedVideoFactory::CreateInstance(cap)->c
> > lo
> > ne()); }
> > 
> > 
> > 
> > Thanks -- Iurii Gordiienko