[Live-devel] Infinite loop in liveMedia/MediaSession.cpp with Windows CE

[Denis Charmet]

Ross Finlayson wrote:
>> I agree that WinCE should not be doing that. However, IMHO, requesting
>> a port over and over, assuming that the OS will eventually give you
>> something suitable, is a risky behavior
>
> There's really no alternative.  We can't just arbitrarily choose an 
> even number ourselves, in case we get a port number that happens to be 
> used by someone else.  And we also have to make sure that - if the 
> code is run in more than one process (i.e., application) on the same 
> host - we don't end up with more than one process accidentally using 
> the same port number.
>
>
>> So I think that it would be appropriate for live555 code to implement
>> some measures making sure that this doesn't happen, and while we're at
>> it, possibly working around OS bugs.
>
> No, in general we're not going to "work around OS bugs" - especially 
> ones as aggregious as this.  What I will do, however, is change the 
> code - where it "Add()"s a record for a port number to the hash table 
> - to check whether there was already a hash table entry (for the same 
> port number) there - so at least we don't end up with a memory leak in 
> this situation.
There is also the 'failure' memory leak:

if (!success) break;  // a fatal error occurred trying to create the RTP 
and RTCP sockets; we can't continue    (liveMedia/MediaSession.cpp : 685)

which IMHO could become:

if (!success) {
  Groupsock* oldGS;
  while ((oldGS = (Groupsock*)socketHashTable->RemoveNext()) != NULL) {
    delete oldGS;
  }
  delete socketHashTable;
  break; // a fatal error occurred trying to create the RTP and RTCP 
sockets; we can't continue
}


By the way Microsoft's excuse for that is that getsockname can return 
false informations before the socket is connected as devices can be 
multihomed.
http://msdn.microsoft.com/en-us/library/ms738543(VS.85).aspx


Thanks for the answers.

Best regards.

-- 
Denis Charmet