Re: freeing CORBA arguments

[<dahaverk rockwellcollins com>]

> Hi Diego,
>
> On 5 Sep 2001, Diego González wrote:

> > On Wed, 2001-09-05 at 10:41, Mark McLoughlin wrote:
>> > Hi Diego,
>> >
>> > On 4 Sep 2001, Diego González wrote:
>> >
>> > >    i want to avoid duplicating them, is there any way to do that?
>> >
>> >      No, fortunately, there isn't ;)
>>
>> i have taken a look at the CORBA C mapping spec from OMG,
>> is it possible to use CORBA_*_set_release functions to avoid
>> having to free the values?
>
>    This is not what these functions are intended for. Basically,
> if a sequence or any has a _buffer that has been allocated from the
> heap, this should be  set to TRUE.
>
>    The fact is - if you don't want the data allocated by the ORB
> for the duration of a method invocation to go away - you need to 'dup'
> it. Trying to get around this is playing with fire. Why is it so
> important to you to circumvent this behaviour?
>
> Cheers,
> Mark

I'd say performance is a reason to try to make memory allocations as static
as possible.

Take a look at this presentation that OIS made at one of the High
performance computing meetings.
The issues he points out are very familiar to people that develop for
embedded systems.


http://www.ois.com/technical/default.htm   <--- Toplevel

http://www.ois.com/technical/ORB_Performance_Enemies__Writing_Fast_IDL.pdf


BTW,  We do have the DII COE test that are mentioned on the OIS web site
working with ORBit.   I'm looking forward to seeing if the DII COE test
show that ORBit2 is an improvement.
I am also trying to "borrow" an OIS ORBexpress license from our Software
Radio people so we can compare it to ORBit.

Later this fall, I plan to get my management's OK to post the graphs of our
comparisons.  (i.e. ACE TAO, MICO, ORBit).


David Haverkamp